Tag Archives: workflow

Automating Amazon EBS Snapshot Management with AWS Step Functions and Amazon CloudWatch Events

Post Syndicated from Andy Katz original https://aws.amazon.com/blogs/compute/automating-amazon-ebs-snapshot-management-with-aws-step-functions-and-amazon-cloudwatch-events/

Brittany Doncaster, Solutions Architect

Business continuity is important for building mission-critical workloads on AWS. As an AWS customer, you might define recovery point objectives (RPO) and recovery time objectives (RTO) for different tier applications in your business. After the RPO and RTO requirements are defined, it is up to your architects to determine how to meet those requirements.

You probably store persistent data in Amazon EBS volumes, which live within a single Availability Zone. And, following best practices, you take snapshots of your EBS volumes to back up the data on Amazon S3, which provides 11 9’s of durability. If you are following these best practices, then you’ve probably recognized the need to manage the number of snapshots you keep for a particular EBS volume and delete older, unneeded snapshots. Doing this cleanup helps save on storage costs.

Some customers also have policies stating that backups need to be stored a certain number of miles away as part of a disaster recovery (DR) plan. To meet these requirements, customers copy their EBS snapshots to the DR region. Then, the same snapshot management and cleanup has to also be done in the DR region.

All of this snapshot management logic consists of different components. You would first tag your snapshots so you could manage them. Then, determine how many snapshots you currently have for a particular EBS volume and assess that value against a retention rule. If the number of snapshots was greater than your retention value, then you would clean up old snapshots. And finally, you might copy the latest snapshot to your DR region. All these steps are just an example of a simple snapshot management workflow. But how do you automate something like this in AWS? How do you do it without servers?

One of the most powerful AWS services released in 2016 was Amazon CloudWatch Events. It enables you to build event-driven IT automation, based on events happening within your AWS infrastructure. CloudWatch Events integrates with AWS Lambda to let you execute your custom code when one of those events occurs. However, the actions to take based on those events aren’t always composed of a single Lambda function. Instead, your business logic may consist of multiple steps (like in the case of the example snapshot management flow described earlier). And you may want to run those steps in sequence or in parallel. You may also want to have retry logic or exception handling for each step.

AWS Step Functions serves just this purpose―to help you coordinate your functions and microservices. Step Functions enables you to simplify your effort and pull the error handling, retry logic, and workflow logic out of your Lambda code. Step Functions integrates with Lambda to provide a mechanism for building complex serverless applications. Now, you can kick off a Step Functions state machine based on a CloudWatch event.

In this post, I discuss how you can target Step Functions in a CloudWatch Events rule. This allows you to have event-driven snapshot management based on snapshot completion events firing in CloudWatch Event rules.

As an example of what you could do with Step Functions and CloudWatch Events, we’ve developed a reference architecture that performs management of your EBS snapshots.

Automating EBS Snapshot Management with Step Functions

This architecture assumes that you have already set up CloudWatch Events to create the snapshots on a schedule or that you are using some other means of creating snapshots according to your needs.

This architecture covers the pieces of the workflow that need to happen after a snapshot has been created.

  • It creates a CloudWatch Events rule to invoke a Step Functions state machine execution when an EBS snapshot is created.
  • The state machine then tags the snapshot, cleans up the oldest snapshots if the number of snapshots is greater than the defined number to retain, and copies the snapshot to a DR region.
  • When the DR region snapshot copy is completed, another state machine kicks off in the DR region. The new state machine has a similar flow and uses some of the same Lambda code to clean up the oldest snapshots that are greater than the defined number to retain.
  • Also, both state machines demonstrate how you can use Step Functions to handle errors within your workflow. Any errors that are caught during execution result in the execution of a Lambda function that writes a message to an SNS topic. Therefore, if any errors occur, you can subscribe to the SNS topic and get notified.

The following is an architecture diagram of the reference architecture:

Creating the Lambda functions and Step Functions state machines

First, pull the code from GitHub and use the AWS CLI to create S3 buckets for the Lambda code in the primary and DR regions. For this example, assume that the primary region is us-west-2 and the DR region is us-east-2. Run the following commands, replacing the italicized text in <> with your own unique bucket names.

git clone https://github.com/awslabs/aws-step-functions-ebs-snapshot-mgmt.git

cd aws-step-functions-ebs-snapshot-mgmt/

aws s3 mb s3://<primary region bucket name> --region us-west-2

aws s3 mb s3://<DR region bucket name> --region us-east-2

Next, use the Serverless Application Model (SAM), which uses AWS CloudFormation to deploy the Lambda functions and Step Functions state machines in the primary and DR regions. Replace the italicized text in <> with the S3 bucket names that you created earlier.

aws cloudformation package --template-file PrimaryRegionTemplate.yaml --s3-bucket <primary region bucket name>  --output-template-file tempPrimary.yaml --region us-west-2

aws cloudformation deploy --template-file tempPrimary.yaml --stack-name ebsSnapshotMgmtPrimary --capabilities CAPABILITY_IAM --region us-west-2

aws cloudformation package --template-file DR_RegionTemplate.yaml --s3-bucket <DR region bucket name> --output-template-file tempDR.yaml  --region us-east-2

aws cloudformation deploy --template-file tempDR.yaml --stack-name ebsSnapshotMgmtDR --capabilities CAPABILITY_IAM --region us-east-2

CloudWatch event rule verification

The CloudFormation templates deploy the following resources:

  • The Lambda functions that are coordinated by Step Functions
  • The Step Functions state machine
  • The SNS topic
  • The CloudWatch Events rules that trigger the state machine execution

So, all of the CloudWatch event rules have been created for you by performing the preceding commands. The next section demonstrates how you could create the CloudWatch event rule manually. To jump straight to testing the workflow, see the “Testing in your Account” section. Otherwise, you begin by setting up the CloudWatch event rule in the primary region for the createSnapshot event and also the CloudWatch event rule in the DR region for the copySnapshot command.

First, open the CloudWatch console in the primary region.

Choose Create Rule and create a rule for the createSnapshot command, with your newly created Step Function state machine as the target.

For Event Source, choose Event Pattern and specify the following values:

  • Service Name: EC2
  • Event Type: EBS Snapshot Notification
  • Specific Event: createSnapshot

For Target, choose Step Functions state machine, then choose the state machine created by the CloudFormation commands. Choose Create a new role for this specific resource. Your completed rule should look like the following:

Choose Configure Details and give the rule a name and description.

Choose Create Rule. You now have a CloudWatch Events rule that triggers a Step Functions state machine execution when the EBS snapshot creation is complete.

Now, set up the CloudWatch Events rule in the DR region as well. This looks almost same, but is based off the copySnapshot event instead of createSnapshot.

In the upper right corner in the console, switch to your DR region. Choose CloudWatch, Create Rule.

For Event Source, choose Event Pattern and specify the following values:

  • Service Name: EC2
  • Event Type: EBS Snapshot Notification
  • Specific Event: copySnapshot

For Target, choose Step Functions state machine, then select the state machine created by the CloudFormation commands. Choose Create a new role for this specific resource. Your completed rule should look like in the following:

As in the primary region, choose Configure Details and then give this rule a name and description. Complete the creation of the rule.

Testing in your account

To test this setup, open the EC2 console and choose Volumes. Select a volume to snapshot. Choose Actions, Create Snapshot, and then create a snapshot.

This results in a new execution of your state machine in the primary and DR regions. You can view these executions by going to the Step Functions console and selecting your state machine.

From there, you can see the execution of the state machine.

Primary region state machine:

DR region state machine:

I’ve also provided CloudFormation templates that perform all the earlier setup without using git clone and running the CloudFormation commands. Choose the Launch Stack buttons below to launch the primary and DR region stacks in Dublin and Ohio, respectively. From there, you can pick up at the Testing in Your Account section above to finish the example. All of the code for this example architecture is located in the aws-step-functions-ebs-snapshot-mgmt AWSLabs repo.

Launch EBS Snapshot Management into Ireland with CloudFormation
Primary Region eu-west-1 (Ireland)

Launch EBS Snapshot Management into Ohio with CloudFormation
DR Region us-east-2 (Ohio)

Summary

This reference architecture is just an example of how you can use Step Functions and CloudWatch Events to build event-driven IT automation. The possibilities are endless:

  • Use this pattern to perform other common cleanup type jobs such as managing Amazon RDS snapshots, old versions of Lambda functions, or old Amazon ECR images—all triggered by scheduled events.
  • Use Trusted Advisor events to identify unused EC2 instances or EBS volumes, then coordinate actions on them, such as alerting owners, stopping, or snapshotting.

Happy coding and please let me know what useful state machines you build!

Automate Your IT Operations Using AWS Step Functions and Amazon CloudWatch Events

Post Syndicated from Andy Katz original https://aws.amazon.com/blogs/compute/automate-your-it-operations-using-aws-step-functions-and-amazon-cloudwatch-events/


Rob Percival, Associate Solutions Architect

Are you interested in reducing the operational overhead of your AWS Cloud infrastructure? One way to achieve this is to automate the response to operational events for resources in your AWS account.

Amazon CloudWatch Events provides a near real-time stream of system events that describe the changes and notifications for your AWS resources. From this stream, you can create rules to route specific events to AWS Step Functions, AWS Lambda, and other AWS services for further processing and automated actions.

In this post, learn how you can use Step Functions to orchestrate serverless IT automation workflows in response to CloudWatch events sourced from AWS Health, a service that monitors and generates events for your AWS resources. As a real-world example, I show automating the response to a scenario where an IAM user access key has been exposed.

Serverless workflows with Step Functions and Lambda

Step Functions makes it easy to develop and orchestrate components of operational response automation using visual workflows. Building automation workflows from individual Lambda functions that perform discrete tasks lets you develop, test, and modify the components of your workflow quickly and seamlessly. As serverless services, Step Functions and Lambda also provide the benefits of more productive development, reduced operational overhead, and no costs incurred outside of when the workflows are actively executing.

Example workflow

As an example, this post focuses on automating the response to an event generated by AWS Health when an IAM access key has been publicly exposed on GitHub. This is a diagram of the automation workflow:

AWS proactively monitors popular code repository sites for IAM access keys that have been publicly exposed. Upon detection of an exposed IAM access key, AWS Health generates an AWS_RISK_CREDENTIALS_EXPOSED event in the AWS account related to the exposed key. A configured CloudWatch Events rule detects this event and invokes a Step Functions state machine. The state machine then orchestrates the automated workflow that deletes the exposed IAM access key, summarizes the recent API activity for the exposed key, and sends the summary message to an Amazon SNS topic to notify the subscribers―in that order.

The corresponding Step Functions state machine diagram of this automation workflow can be seen below:

While this particular example focuses on IT automation workflows in response to the AWS_RISK_CREDENTIALS_EXPOSEDevent sourced from AWS Health, it can be generalized to integrate with other events from these services, other event-generating AWS services, and even run on a time-based schedule.

Walkthrough

To follow along, use the code and resources found in the aws-health-tools GitHub repo. The code and resources include an AWS CloudFormation template, in addition to instructions on how to use it.

Launch Stack into N. Virginia with CloudFormation

The Step Functions state machine execution starts with the exposed keys event details in JSON, a sanitized example of which is provided below:

{
    "version": "0",
    "id": "121345678-1234-1234-1234-123456789012",
    "detail-type": "AWS Health Event",
    "source": "aws.health",
    "account": "123456789012",
    "time": "2016-06-05T06:27:57Z",
    "region": "us-east-1",
    "resources": [],
    "detail": {
        "eventArn": "arn:aws:health:us-east-1::event/AWS_RISK_CREDENTIALS_EXPOSED_XXXXXXXXXXXXXXXXX",
        "service": "RISK",
        "eventTypeCode": "AWS_RISK_CREDENTIALS_EXPOSED",
        "eventTypeCategory": "issue",
        "startTime": "Sat, 05 Jun 2016 15:10:09 GMT",
        "eventDescription": [
            {
                "language": "en_US",
                "latestDescription": "A description of the event is provided here"
            }
        ],
        "affectedEntities": [
            {
                "entityValue": "ACCESS_KEY_ID_HERE"
            }
        ]
    }
}

After it’s invoked, the state machine execution proceeds as follows.

Step 1: Delete the exposed IAM access key pair

The first thing you want to do when you determine that an IAM access key has been exposed is to delete the key pair so that it can no longer be used to make API calls. This Step Functions task state deletes the exposed access key pair detailed in the incoming event, and retrieves the IAM user associated with the key to look up API activity for the user in the next step. The user name, access key, and other details about the event are passed to the next step as JSON.

This state contains a powerful error-handling feature offered by Step Functions task states called a catch configuration. Catch configurations allow you to reroute and continue state machine invocation at new states depending on potential errors that occur in your task function. In this case, the catch configuration skips to Step 3. It immediately notifies your security team that errors were raised in the task function of this step (Step 1), when attempting to look up the corresponding IAM user for a key or delete the user’s access key.

Note: Step Functions also offers a retry configuration for when you would rather retry a task function that failed due to error, with the option to specify an increasing time interval between attempts and a maximum number of attempts.

Step 2: Summarize recent API activity for key

After you have deleted the access key pair, you’ll want to have some immediate insight into whether it was used for malicious activity in your account. Another task state, this step uses AWS CloudTrail to look up and summarize the most recent API activity for the IAM user associated with the exposed key. The summary is in the form of counts for each API call made and resource type and name affected. This summary information is then passed to the next step as JSON. This step requires information that you obtained in Step 1. Step Functions ensures the successful completion of Step 1 before moving to Step 2.

Step 3: Notify security

The summary information gathered in the last step can provide immediate insight into any malicious activity on your account made by the exposed key. To determine this and further secure your account if necessary, you must notify your security team with the gathered summary information.

This final task state generates an email message providing in-depth detail about the event using the API activity summary, and publishes the message to an SNS topic subscribed to by the members of your security team.

If the catch configuration of the task state in Step 1 was triggered, then the security notification email instead directs your security team to log in to the console and navigate to the Personal Health Dashboard to view more details on the incident.

Lessons learned

When implementing this use case with Step Functions and Lambda, consider the following:

  • One of the most important parts of implementing automation in response to operational events is to ensure visibility into the response and resolution actions is retained. Step Functions and Lambda enable you to orchestrate your granular response and resolution actions that provides direct visibility into the state of the automation workflow.
  • This basic workflow currently executes these steps serially with a catch configuration for error handling. More sophisticated workflows can leverage the parallel execution, branching logic, and time delay functionality provided by Step Functions.
  • Catch and retry configurations for task states allow for orchestrating reliable workflows while maintaining the granularity of each Lambda function. Without leveraging a catch configuration in Step 1, you would have had to duplicate code from the function in Step 3 to ensure that your security team was notified on failure to delete the access key.
  • Step Functions and Lambda are serverless services, so there is no cost for these services when they are not running. Because this IT automation workflow only runs when an IAM access key is exposed for this account (which is hopefully rare!), the total monthly cost for this workflow is essentially $0.

Conclusion

Automating the response to operational events for resources in your AWS account can free up the valuable time of your engineers. Step Functions and Lambda enable granular IT automation workflows to achieve this result while gaining direct visibility into the orchestration and state of the automation.

For more examples of how to use Step Functions to automate the operations of your AWS resources, or if you’d like to see how Step Functions can be used to build and orchestrate serverless applications, visit Getting Started on the Step Functions website.

Delivering Graphics Apps with Amazon AppStream 2.0

Post Syndicated from Deepak Suryanarayanan original https://aws.amazon.com/blogs/compute/delivering-graphics-apps-with-amazon-appstream-2-0/

Sahil Bahri, Sr. Product Manager, Amazon AppStream 2.0

Do you need to provide a workstation class experience for users who run graphics apps? With Amazon AppStream 2.0, you can stream graphics apps from AWS to a web browser running on any supported device. AppStream 2.0 offers a choice of GPU instance types. The range includes the newly launched Graphics Design instance, which allows you to offer a fast, fluid user experience at a fraction of the cost of using a graphics workstation, without upfront investments or long-term commitments.

In this post, I discuss the Graphics Design instance type in detail, and how you can use it to deliver a graphics application such as Siemens NX―a popular CAD/CAM application that we have been testing on AppStream 2.0 with engineers from Siemens PLM.

Graphics Instance Types on AppStream 2.0

First, a quick recap on the GPU instance types available with AppStream 2.0. In July, 2017, we launched graphics support for AppStream 2.0 with two new instance types that Jeff Barr discussed on the AWS Blog:

  • Graphics Desktop
  • Graphics Pro

Many customers in industries such as engineering, media, entertainment, and oil and gas are using these instances to deliver high-performance graphics applications to their users. These instance types are based on dedicated NVIDIA GPUs and can run the most demanding graphics applications, including those that rely on CUDA graphics API libraries.

Last week, we added a new lower-cost instance type: Graphics Design. This instance type is a great fit for engineers, 3D modelers, and designers who use graphics applications that rely on the hardware acceleration of DirectX, OpenGL, or OpenCL APIs, such as Siemens NX, Autodesk AutoCAD, or Adobe Photoshop. The Graphics Design instance is based on AMD’s FirePro S7150x2 Server GPUs and equipped with AMD Multiuser GPU technology. The instance type uses virtualized GPUs to achieve lower costs, and is available in four instance sizes to scale and match the requirements of your applications.

Instance vCPUs Instance RAM (GiB) GPU Memory (GiB)
stream.graphics-design.large 2 7.5 GiB 1
stream.graphics-design.xlarge 4 15.3 GiB 2
stream.graphics-design.2xlarge 8 30.5 GiB 4
stream.graphics-design.4xlarge 16 61 GiB 8

The following table compares all three graphics instance types on AppStream 2.0, along with example applications you could use with each.

  Graphics Design Graphics Desktop Graphics Pro
Number of instance sizes 4 1 3
GPU memory range
1–8 GiB 4 GiB 8–32 GiB
vCPU range 2–16 8 16–32
Memory range 7.5–61 GiB 15 GiB 122–488 GiB
Graphics libraries supported AMD FirePro S7150x2 NVIDIA GRID K520 NVIDIA Tesla M60
Price range (N. Virginia AWS Region) $0.25 – $2.00/hour $0.5/hour $2.05 – $8.20/hour
Example applications Adobe Premiere Pro, AutoDesk Revit, Siemens NX AVEVA E3D, SOLIDWORKS AutoDesk Maya, Landmark DecisionSpace, Schlumberger Petrel

Example graphics instance set up with Siemens NX

In the section, I walk through setting up Siemens NX with Graphics Design instances on AppStream 2.0. After set up is complete, users can able to access NX from within their browser and also access their design files from a file share. You can also use these steps to set up and test your own graphics applications on AppStream 2.0. Here’s the workflow:

  1. Create a file share to load and save design files.
  2. Create an AppStream 2.0 image with Siemens NX installed.
  3. Create an AppStream 2.0 fleet and stack.
  4. Invite users to access Siemens NX through a browser.
  5. Validate the setup.

To learn more about AppStream 2.0 concepts and set up, see the previous post Scaling Your Desktop Application Streams with Amazon AppStream 2.0. For a deeper review of all the setup and maintenance steps, see Amazon AppStream 2.0 Developer Guide.

Step 1: Create a file share to load and save design files

To launch and configure the file server

  1. Open the EC2 console and choose Launch Instance.
  2. Scroll to the Microsoft Windows Server 2016 Base Image and choose Select.
  3. Choose an instance type and size for your file server (I chose the general purpose m4.large instance). Choose Next: Configure Instance Details.
  4. Select a VPC and subnet. You launch AppStream 2.0 resources in the same VPC. Choose Next: Add Storage.
  5. If necessary, adjust the size of your EBS volume. Choose Review and Launch, Launch.
  6. On the Instances page, give your file server a name, such as My File Server.
  7. Ensure that the security group associated with the file server instance allows for incoming traffic from the security group that you select for your AppStream 2.0 fleets or image builders. You can use the default security group and select the same group while creating the image builder and fleet in later steps.

Log in to the file server using a remote access client such as Microsoft Remote Desktop. For more information about connecting to an EC2 Windows instance, see Connect to Your Windows Instance.

To enable file sharing

  1. Create a new folder (such as C:\My Graphics Files) and upload the shared files to make available to your users.
  2. From the Windows control panel, enable network discovery.
  3. Choose Server Manager, File and Storage Services, Volumes.
  4. Scroll to Shares and choose Start the Add Roles and Features Wizard. Go through the wizard to install the File Server and Share role.
  5. From the left navigation menu, choose Shares.
  6. Choose Start the New Share Wizard to set up your folder as a file share.
  7. Open the context (right-click) menu on the share and choose Properties, Permissions, Customize Permissions.
  8. Choose Permissions, Add. Add Read and Execute permissions for everyone on the network.

Step 2:  Create an AppStream 2.0 image with Siemens NX installed

To connect to the image builder and install applications

  1. Open the AppStream 2.0 management console and choose Images, Image Builder, Launch Image Builder.
  2. Create a graphics design image builder in the same VPC as your file server.
  3. From the Image builder tab, select your image builder and choose Connect. This opens a new browser tab and display a desktop to log in to.
  4. Log in to your image builder as ImageBuilderAdmin.
  5. Launch the Image Assistant.
  6. Download and install Siemens NX and other applications on the image builder. I added Blender and Firefox, but you could replace these with your own applications.
  7. To verify the user experience, you can test the application performance on the instance.

Before you finish creating the image, you must mount the file share by enabling a few Microsoft Windows services.

To mount the file share

  1. Open services.msc and check the following services:
  • DNS Client
  • Function Discovery Resource Publication
  • SSDP Discovery
  • UPnP Device H
  1. If any of the preceding services have Startup Type set to Manual, open the context (right-click) menu on the service and choose Start. Otherwise, open the context (right-click) menu on the service and choose Properties. For Startup Type, choose Manual, Apply. To start the service, choose Start.
  2. From the Windows control panel, enable network discovery.
  3. Create a batch script that mounts a file share from the storage server set up earlier. The file share is mounted automatically when a user connects to the AppStream 2.0 environment.

Logon Script Location: C:\Users\Public\logon.bat

Script Contents:

:loop

net use H: \\path\to\network\share 

PING localhost -n 30 >NUL

IF NOT EXIST H:\ GOTO loop

  1. Open gpedit.msc and choose User Configuration, Windows Settings, Scripts. Set logon.bat as the user logon script.
  2. Next, create a batch script that makes the mounted drive visible to the user.

Logon Script Location: C:\Users\Public\startup.bat

Script Contents:
REG DELETE “HKEY_LOCAL_MACHINE\Software\Microsoft\Windows\CurrentVersion\Policies\Explorer” /v “NoDrives” /f

  1. Open Task Scheduler and choose Create Task.
  2. Choose General, provide a task name, and then choose Change User or Group.
  3. For Enter the object name to select, enter SYSTEM and choose Check Names, OK.
  4. Choose Triggers, New. For Begin the task, choose At startup. Under Advanced Settings, change Delay task for to 5 minutes. Choose OK.
  5. Choose Actions, New. Under Settings, for Program/script, enter C:\Users\Public\startup.bat. Choose OK.
  6. Choose Conditions. Under Power, clear the Start the task only if the computer is on AC power Choose OK.
  7. To view your scheduled task, choose Task Scheduler Library. Close Task Scheduler when you are done.

Step 3:  Create an AppStream 2.0 fleet and stack

To create a fleet and stack

  1. In the AppStream 2.0 management console, choose Fleets, Create Fleet.
  2. Give the fleet a name, such as Graphics-Demo-Fleet, that uses the newly created image and the same VPC as your file server.
  3. Choose Stacks, Create Stack. Give the stack a name, such as Graphics-Demo-Stack.
  4. After the stack is created, select it and choose Actions, Associate Fleet. Associate the stack with the fleet you created in step 1.

Step 4:  Invite users to access Siemens NX through a browser

To invite users

  1. Choose User Pools, Create User to create users.
  2. Enter a name and email address for each user.
  3. Select the users just created, and choose Actions, Assign Stack to provide access to the stack created in step 2. You can also provide access using SAML 2.0 and connect to your Active Directory if necessary. For more information, see the Enabling Identity Federation with AD FS 3.0 and Amazon AppStream 2.0 post.

Your user receives an email invitation to set up an account and use a web portal to access the applications that you have included in your stack.

Step 5:  Validate the setup

Time for a test drive with Siemens NX on AppStream 2.0!

  1. Open the link for the AppStream 2.0 web portal shared through the email invitation. The web portal opens in your default browser. You must sign in with the temporary password and set a new password. After that, you get taken to your app catalog.
  2. Launch Siemens NX and interact with it using the demo files available in the shared storage folder – My Graphics Files. 

After I launched NX, I captured the screenshot below. The Siemens PLM team also recorded a video with NX running on AppStream 2.0.

Summary

In this post, I discussed the GPU instances available for delivering rich graphics applications to users in a web browser. While I demonstrated a simple setup, you can scale this out to launch a production environment with users signing in using Active Directory credentials,  accessing persistent storage with Amazon S3, and using other commonly requested features reviewed in the Amazon AppStream 2.0 Launch Recap – Domain Join, Simple Network Setup, and Lots More post.

To learn more about AppStream 2.0 and capabilities added this year, see Amazon AppStream 2.0 Resources.

A Hardware Privacy Monitor for iPhones

Post Syndicated from Bruce Schneier original https://www.schneier.com/blog/archives/2017/09/a_hardware_priv.html

Andrew “bunnie” Huang and Edward Snowden have designed a hardware device that attaches to an iPhone and monitors it for malicious surveillance activities, even in instances where the phone’s operating system has been compromised. They call it an Introspection Engine, and their use model is a journalist who is concerned about government surveillance:

Our introspection engine is designed with the following goals in mind:

  1. Completely open source and user-inspectable (“You don’t have to trust us”)
  2. Introspection operations are performed by an execution domain completely separated from the phone”s CPU (“don’t rely on those with impaired judgment to fairly judge their state”)

  3. Proper operation of introspection system can be field-verified (guard against “evil maid” attacks and hardware failures)

  4. Difficult to trigger a false positive (users ignore or disable security alerts when there are too many positives)

  5. Difficult to induce a false negative, even with signed firmware updates (“don’t trust the system vendor” — state-level adversaries with full cooperation of system vendors should not be able to craft signed firmware updates that spoof or bypass the introspection engine)

  6. As much as possible, the introspection system should be passive and difficult to detect by the phone’s operating system (prevent black-listing/targeting of users based on introspection engine signatures)

  7. Simple, intuitive user interface requiring no specialized knowledge to interpret or operate (avoid user error leading to false negatives; “journalists shouldn’t have to be cryptographers to be safe”)

  8. Final solution should be usable on a daily basis, with minimal impact on workflow (avoid forcing field reporters into the choice between their personal security and being an effective journalist)

This looks like fantastic work, and they have a working prototype.

Of course, this does nothing to stop all the legitimate surveillance that happens over a cell phone: location tracking, records of who you talk to, and so on.

BoingBoing post.

Implement Continuous Integration and Delivery of Apache Spark Applications using AWS

Post Syndicated from Luis Caro Perez original https://aws.amazon.com/blogs/big-data/implement-continuous-integration-and-delivery-of-apache-spark-applications-using-aws/

When you develop Apache Spark–based applications, you might face some additional challenges when dealing with continuous integration and deployment pipelines, such as the following common issues:

  • Applications must be tested on real clusters using automation tools (live test)
  • Any user or developer must be able to easily deploy and use different versions of both the application and infrastructure to be able to debug, experiment on, and test different functionality.
  • Infrastructure needs to be evaluated and tested along with the application that uses it.

In this post, we walk you through a solution that implements a continuous integration and deployment pipeline supported by AWS services. The pipeline offers the following workflow:

  • Deploy the application to a QA stage after a commit is performed to the source code.
  • Perform a unit test using Spark local mode.
  • Deploy to a dynamically provisioned Amazon EMR cluster and test the Spark application on it
  • Update the application as an AWS Service Catalog product version, allowing a user to deploy any version (commit) of the application on demand.

Solution overview

The following diagram shows the pipeline workflow.

The solution uses AWS CodePipeline, which allows users to orchestrate and automate the build, test, and deploy stages for application source code. The solution consists of a pipeline that contains the following stages:

  • Source: Both the Spark application source code in addition to the AWS CloudFormation template file for deploying the application are committed to version control. In this example, we use AWS CodeCommit. For an example of the application source code, see zip. 
  • Build: In this stage, you use Apache Maven both to generate the application .jar binaries and to execute all of the application unit tests that end with *Spec.scala. In this example, we use AWS CodeBuild, which runs the unit tests given that they are designed to use Spark local mode.
  • QADeploy: In this stage, the .jar file built previously is deployed using the CloudFormation template included with the application source code. All the resources are created in this stage, such as networks, EMR clusters, and so on. 
  • LiveTest: In this stage, you use Apache Maven to execute all the application tests that end with *SpecLive.scala. The tests submit EMR steps to the cluster created as part of the QADeploy step. The tests verify that the steps ran successfully and their results. 
  • LiveTestApproval: This stage is included in case a pipeline administrator approval is required to deploy the application to the next stages. The pipeline pauses in this stage until an administrator manually approves the release. 
  • QACleanup: In this stage, you use an AWS Lambda function to delete the CloudFormation template deployed as part of the QADeploy stage. The function does not affect any resources other than those deployed as part of the QADeploy stage. 
  • DeployProduct: In this stage, you use a Lambda function that creates or updates an AWS Service Catalog product and portfolio. Every time the pipeline releases a change to the application, the AWS Service Catalog product gets a new version, with the commit of the change as the version description. 

Try it out!

Use the provided sample template to get started using this solution. This template creates the pipeline described earlier with all of its stages. It performs an initial commit of the sample Spark application in order to trigger the first release change. To deploy the template, use the following AWS CLI command:

aws cloudformation create-stack  --template-url https://s3.amazonaws.com/aws-bigdata-blog/artifacts/sparkAppDemoForPipeline/emrSparkpipeline.yaml --stack-name emr-spark-pipeline --capabilities CAPABILITY_NAMED_IAM

After the template finishes creating resources, you see the pipeline name on the stack Outputs tab. After that, open the AWS CodePipeline console and select the newly created pipeline.

After a couple of minutes, AWS CodePipeline detects the initial commit applied by the CloudFormation stack and starts the first release.

You can watch how the pipeline goes through the Build, QADeploy, and LiveTest stages until it finally reaches the LiveTestApproval stage.

At this point, you can check the results of the test in the log files of the Build and LiveTest stage jobs on AWS CodeBuild. If you check the CloudFormation console, you see that a new template has been deployed as part of the QADeploy stage.

You can also visit the EMR console and view how the LiveTest stage submitted steps to the EMR cluster.

After performing the review, manually approve the revision on the LiveTestApproval stage by using the AWS CodePipeline console.

After the revision is approved, the pipeline proceeds to use a Lambda function that destroys the resources deployed on the QAdeploy stage. Finally, it creates or updates a product and portfolio in AWS Service Catalog. After the final stage of the pipeline is complete, you can check that the product is created successfully on the AWS Service Catalog console.

You can check the product versions and notice that the first version is the initial commit performed by the CloudFormation template.

You can proceed to share the created portfolio with any users in your AWS account and allow them to deploy any version of the Spark application. You can also perform a commit on the AWS CodeCommit repository. The pipeline is triggered automatically and repeats the pipeline execution to deploy a new version of the product.

To destroy all of the resources created by the stack, make sure all the deployed stacks using AWS Service Catalog or the QAdeploy stage are destroyed. Then, destroy the pipeline template using the following AWS CLI command:

 

aws cloudformation delete-stack --stack-name emr-spark-pipeline

Conclusion

You can use the sample template and Spark application shared in this post and adapt them for the specific needs of your own application. The pipeline can have as many stages as needed and it can be used to automatically deploy to AWS Service Catalog or a production environment using CloudFormation.

If you have questions or suggestions, please comment below.


Additional Reading

Learn how to implement authorization and auditing on Amazon EMR using Apache Ranger.

 


About the Authors

Luis Caro is a Big Data Consultant for AWS Professional Services. He works with our customers to provide guidance and technical assistance on big data projects, helping them improving the value of their solutions when using AWS.

 

 

Samuel Schmidt is a Big Data Consultant for AWS Professional Services. He works with our customers to provide guidance and technical assistance on big data projects, helping them improving the value of their solutions when using AWS.

 

 

 

AWS Hot Startups – August 2017

Post Syndicated from Tina Barr original https://aws.amazon.com/blogs/aws/aws-hot-startups-august-2017/

There’s no doubt about it – Artificial Intelligence is changing the world and how it operates. Across industries, organizations from startups to Fortune 500s are embracing AI to develop new products, services, and opportunities that are more efficient and accessible for their consumers. From driverless cars to better preventative healthcare to smart home devices, AI is driving innovation at a fast rate and will continue to play a more important role in our everyday lives.

This month we’d like to highlight startups using AI solutions to help companies grow. We are pleased to feature:

  • SignalBox – a simple and accessible deep learning platform to help businesses get started with AI.
  • Valossa – an AI video recognition platform for the media and entertainment industry.
  • Kaliber – innovative applications for businesses using facial recognition, deep learning, and big data.

SignalBox (UK)

In 2016, SignalBox founder Alain Richardt was hearing the same comments being made by developers, data scientists, and business leaders. They wanted to get into deep learning but didn’t know where to start. Alain saw an opportunity to commodify and apply deep learning by providing a platform that does the heavy lifting with an easy-to-use web interface, blueprints for common tasks, and just a single-click to productize the models. With SignalBox, companies can start building deep learning models with no coding at all – they just select a data set, choose a network architecture, and go. SignalBox also offers step-by-step tutorials, tips and tricks from industry experts, and consulting services for customers that want an end-to-end AI solution.

SignalBox offers a variety of solutions that are being used across many industries for energy modeling, fraud detection, customer segmentation, insurance risk modeling, inventory prediction, real estate prediction, and more. Existing data science teams are using SignalBox to accelerate their innovation cycle. One innovative UK startup, Energi Mine, recently worked with SignalBox to develop deep networks that predict anomalous energy consumption patterns and do time series predictions on energy usage for businesses with hundreds of sites.

SignalBox uses a variety of AWS services including Amazon EC2, Amazon VPC, Amazon Elastic Block Store, and Amazon S3. The ability to rapidly provision EC2 GPU instances has been a critical factor in their success – both in terms of keeping their operational expenses low, as well as speed to market. The Amazon API Gateway has allowed for operational automation, giving SignalBox the ability to control its infrastructure.

To learn more about SignalBox, visit here.

Valossa (Finland)

As students at the University of Oulu in Finland, the Valossa founders spent years doing research in the computer science and AI labs. During that time, the team witnessed how the world was moving beyond text, with video playing a greater role in day-to-day communication. This spawned an idea to use technology to automatically understand what an audience is viewing and share that information with a global network of content producers. Since 2015, Valossa has been building next generation AI applications to benefit the media and entertainment industry and is moving beyond the capabilities of traditional visual recognition systems.

Valossa’s AI is capable of analyzing any video stream. The AI studies a vast array of data within videos and converts that information into descriptive tags, categories, and overviews automatically. Basically, it sees, hears, and understands videos like a human does. The Valossa AI can detect people, visual and auditory concepts, key speech elements, and labels explicit content to make moderating and filtering content simpler. Valossa’s solutions are designed to provide value for the content production workflow, from media asset management to end-user applications for content discovery. AI-annotated content allows online viewers to jump directly to their favorite scenes or search specific topics and actors within a video.

Valossa leverages AWS to deliver the industry’s first complete AI video recognition platform. Using Amazon EC2 GPU instances, Valossa can easily scale their computation capacity based on customer activity. High-volume video processing with GPU instances provides the necessary speed for time-sensitive workflows. The geo-located Availability Zones in EC2 allow Valossa to bring resources close to their customers to minimize network delays. Valossa also uses Amazon S3 for video ingestion and to provide end-user video analytics, which makes managing and accessing media data easy and highly scalable.

To see how Valossa works, check out www.WhatIsMyMovie.com or enable the Alexa Skill, Valossa Movie Finder. To try the Valossa AI, sign up for free at www.valossa.com.

Kaliber (San Francisco, CA)

Serial entrepreneurs Ray Rahman and Risto Haukioja founded Kaliber in 2016. The pair had previously worked in startups building smart cities and online privacy tools, and teamed up to bring AI to the workplace and change the hospitality industry. Our world is designed to appeal to our senses – stores and warehouses have clearly marked aisles, products are colorfully packaged, and we use these designs to differentiate one thing from another. We tell each other apart by our faces, and previously that was something only humans could measure or act upon. Kaliber is using facial recognition, deep learning, and big data to create solutions for business use. Markets and companies that aren’t typically associated with cutting-edge technology will be able to use their existing camera infrastructure in a whole new way, making them more efficient and better able to serve their customers.

Computer video processing is rapidly expanding, and Kaliber believes that video recognition will extend to far more than security cameras and robots. Using the clients’ network of in-house cameras, Kaliber’s platform extracts key data points and maps them to actionable insights using their machine learning (ML) algorithm. Dashboards connect users to the client’s BI tools via the Kaliber enterprise APIs, and managers can view these analytics to improve their real-world processes, taking immediate corrective action with real-time alerts. Kaliber’s Real Metrics are aimed at combining the power of image recognition with ML to ultimately provide a more meaningful experience for all.

Kaliber uses many AWS services, including Amazon Rekognition, Amazon Kinesis, AWS Lambda, Amazon EC2 GPU instances, and Amazon S3. These services have been instrumental in helping Kaliber meet the needs of enterprise customers in record time.

Learn more about Kaliber here.

Thanks for reading and we’ll see you next month!

-Tina

 

From Data Lake to Data Warehouse: Enhancing Customer 360 with Amazon Redshift Spectrum

Post Syndicated from Dylan Tong original https://aws.amazon.com/blogs/big-data/from-data-lake-to-data-warehouse-enhancing-customer-360-with-amazon-redshift-spectrum/

Achieving a 360o-view of your customer has become increasingly challenging as companies embrace omni-channel strategies, engaging customers across websites, mobile, call centers, social media, physical sites, and beyond. The promise of a web where online and physical worlds blend makes understanding your customers more challenging, but also more important. Businesses that are successful in this medium have a significant competitive advantage.

The big data challenge requires the management of data at high velocity and volume. Many customers have identified Amazon S3 as a great data lake solution that removes the complexities of managing a highly durable, fault tolerant data lake infrastructure at scale and economically.

AWS data services substantially lessen the heavy lifting of adopting technologies, allowing you to spend more time on what matters most—gaining a better understanding of customers to elevate your business. In this post, I show how a recent Amazon Redshift innovation, Redshift Spectrum, can enhance a customer 360 initiative.

Customer 360 solution

A successful customer 360 view benefits from using a variety of technologies to deliver different forms of insights. These could range from real-time analysis of streaming data from wearable devices and mobile interactions to historical analysis that requires interactive, on demand queries on billions of transactions. In some cases, insights can only be inferred through AI via deep learning. Finally, the value of your customer data and insights can’t be fully realized until it is operationalized at scale—readily accessible by fleets of applications. Companies are leveraging AWS for the breadth of services that cover these domains, to drive their data strategy.

A number of AWS customers stream data from various sources into a S3 data lake through Amazon Kinesis. They use Kinesis and technologies in the Hadoop ecosystem like Spark running on Amazon EMR to enrich this data. High-value data is loaded into an Amazon Redshift data warehouse, which allows users to analyze and interact with data through a choice of client tools. Redshift Spectrum expands on this analytics platform by enabling Amazon Redshift to blend and analyze data beyond the data warehouse and across a data lake.

The following diagram illustrates the workflow for such a solution.

This solution delivers value by:

  • Reducing complexity and time to value to deeper insights. For instance, an existing data model in Amazon Redshift may provide insights across dimensions such as customer, geography, time, and product on metrics from sales and financial systems. Down the road, you may gain access to streaming data sources like customer-care call logs and website activity that you want to blend in with the sales data on the same dimensions to understand how web and call center experiences maybe correlated with sales performance. Redshift Spectrum can join these dimensions in Amazon Redshift with data in S3 to allow you to quickly gain new insights, and avoid the slow and more expensive alternative of fully integrating these sources with your data warehouse.
  • Providing an additional avenue for optimizing costs and performance. In cases like call logs and clickstream data where volumes could be many TBs to PBs, storing the data exclusively in S3 yields significant cost savings. Interactive analysis on massive datasets may now be economically viable in cases where data was previously analyzed periodically through static reports generated by inexpensive batch processes. In some cases, you can improve the user experience while simultaneously lowering costs. Spectrum is powered by a large-scale infrastructure external to your Amazon Redshift cluster, and excels at scanning and aggregating large volumes of data. For instance, your analysts maybe performing data discovery on customer interactions across millions of consumers over years of data across various channels. On this large dataset, certain queries could be slow if you didn’t have a large Amazon Redshift cluster. Alternatively, you could use Redshift Spectrum to achieve a better user experience with a smaller cluster.

Proof of concept walkthrough

To make evaluation easier for you, I’ve conducted a Redshift Spectrum proof-of-concept (PoC) for the customer 360 use case. For those who want to replicate the PoC, the instructions, AWS CloudFormation templates, and public data sets are available in the GitHub repository.

The remainder of this post is a journey through the project, observing best practices in action, and learning how you can achieve business value. The walkthrough involves:

  • An analysis of performance data from the PoC environment involving queries that demonstrate blending and analysis of data across Amazon Redshift and S3. Observe that great results are achievable at scale.
  • Guidance by example on query tuning, design, and data preparation to illustrate the optimization process. This includes tuning a query that combines clickstream data in S3 with customer and time dimensions in Amazon Redshift, and aggregates ~1.9 B out of 3.7 B+ records in under 10 seconds with a small cluster!
  • Guidance and measurements to help assess deciding between two options: accessing and analyzing data exclusively in Amazon Redshift, or using Redshift Spectrum to access data left in S3.

Stream ingestion and enrichment

The focus of this post isn’t stream ingestion and enrichment on Kinesis and EMR, but be mindful of performance best practices on S3 to ensure good streaming and query performance:

  • Use random object keys: The data files provided for this project are prefixed with SHA-256 hashes to prevent hot partitions. This is important to ensure that optimal request rates to support PUT requests from the incoming stream in addition to certain queries from large Amazon Redshift clusters that could send a large number of parallel GET requests.
  • Micro-batch your data stream: S3 isn’t optimized for small random write workloads. Your datasets should be micro-batched into large files. For instance, the “parquet-1” dataset provided batches >7 million records per file. The optimal file size for Redshift Spectrum is usually in the 100 MB to 1 GB range.

If you have an edge case that may pose scalability challenges, AWS would love to hear about it. For further guidance, talk to your solutions architect.

Environment

The project consists of the following environment:

  • Amazon Redshift cluster: 4 X dc1.large
  • Data:
    • Time and customer dimension tables are stored on all Amazon Redshift nodes (ALL distribution style):
      • The data originates from the DWDATE and CUSTOMER tables in the Star Schema Benchmark
      • The customer table contains attributes for 3 million customers.
      • The time data is at the day-level granularity, and spans 7 years, from the start of 1992 to the end of 1998.
    • The clickstream data is stored in an S3 bucket, and serves as a fact table.
      • Various copies of this dataset in CSV and Parquet format have been provided, for reasons to be discussed later.
      • The data is a modified version of the uservisits dataset from AMPLab’s Big Data Benchmark, which was generated by Intel’s Hadoop benchmark tools.
      • Changes were minimal, so that existing test harnesses for this test can be adapted:
        • Increased the 751,754,869-row dataset 5X to 3,758,774,345 rows.
        • Added surrogate keys to support joins with customer and time dimensions. These keys were distributed evenly across the entire dataset to represents user visits from six customers over seven years.
        • Values for the visitDate column were replaced to align with the 7-year timeframe, and the added time surrogate key.

Queries across the data lake and data warehouse 

Imagine a scenario where a business analyst plans to analyze clickstream metrics like ad revenue over time and by customer, market segment and more. The example below is a query that achieves this effect: 

The query part highlighted in red retrieves clickstream data in S3, and joins the data with the time and customer dimension tables in Amazon Redshift through the part highlighted in blue. The query returns the total ad revenue for three customers over the last three months, along with info on their respective market segment.

Unfortunately, this query takes around three minutes to run, and doesn’t enable the interactive experience that you want. However, there’s a number of performance optimizations that you can implement to achieve the desired performance.

Performance analysis

Two key utilities provide visibility into Redshift Spectrum:

  • EXPLAIN
    Provides the query execution plan, which includes info around what processing is pushed down to Redshift Spectrum. Steps in the plan that include the prefix S3 are executed on Redshift Spectrum. For instance, the plan for the previous query has the step “S3 Seq Scan clickstream.uservisits_csv10”, indicating that Redshift Spectrum performs a scan on S3 as part of the query execution.
  • SVL_S3QUERY_SUMMARY
    Statistics for Redshift Spectrum queries are stored in this table. While the execution plan presents cost estimates, this table stores actual statistics for past query runs.

You can get the statistics of your last query by inspecting the SVL_S3QUERY_SUMMARY table with the condition (query = pg_last_query_id()). Inspecting the previous query reveals that the entire dataset of nearly 3.8 billion rows was scanned to retrieve less than 66.3 million rows. Improving scan selectivity in your query could yield substantial performance improvements.

Partitioning

Partitioning is a key means to improving scan efficiency. In your environment, the data and tables have already been organized, and configured to support partitions. For more information, see the PoC project setup instructions. The clickstream table was defined as:

CREATE EXTERNAL TABLE clickstream.uservisits_csv10
…
PARTITIONED BY(customer int4, visitYearMonth int4)

The entire 3.8 billion-row dataset is organized as a collection of large files where each file contains data exclusive to a particular customer and month in a year. This allows you to partition your data into logical subsets by customer and year/month. With partitions, the query engine can target a subset of files:

  • Only for specific customers
  • Only data for specific months
  • A combination of specific customers and year/months

You can use partitions in your queries. Instead of joining your customer data on the surrogate customer key (that is, c.c_custkey = uv.custKey), the partition key “customer” should be used instead:

SELECT c.c_name, c.c_mktsegment, t.prettyMonthYear, SUM(uv.adRevenue)
…
ON c.c_custkey = uv.customer
…
ORDER BY c.c_name, c.c_mktsegment, uv.yearMonthKey  ASC

This query should run approximately twice as fast as the previous query. If you look at the statistics for this query in SVL_S3QUERY_SUMMARY, you see that only half the dataset was scanned. This is expected because your query is on three out of six customers on an evenly distributed dataset. However, the scan is still inefficient, and you can benefit from using your year/month partition key as well:

SELECT c.c_name, c.c_mktsegment, t.prettyMonthYear, SUM(uv.adRevenue)
…
ON c.c_custkey = uv.customer
…
ON uv.visitYearMonth = t.d_yearmonthnum
…
ORDER BY c.c_name, c.c_mktsegment, uv.visitYearMonth ASC

All joins between the tables are now using partitions. Upon reviewing the statistics for this query, you should observe that Redshift Spectrum scans and returns the exact number of rows, 66,270,117. If you run this query a few times, you should see execution time in the range of 8 seconds, which is a 22.5X improvement on your original query!

Predicate pushdown and storage optimizations 

Previously, I mentioned that Redshift Spectrum performs processing through large-scale infrastructure external to your Amazon Redshift cluster. It is optimized for performing large scans and aggregations on S3. In fact, Redshift Spectrum may even out-perform a medium size Amazon Redshift cluster on these types of workloads with the proper optimizations. There are two important variables to consider for optimizing large scans and aggregations:

  • File size and count. As a general rule, use files 100 MB-1 GB in size, as Redshift Spectrum and S3 are optimized for reading this object size. However, the number of files operating on a query is directly correlated with the parallelism achievable by a query. There is an inverse relationship between file size and count: the bigger the files, the fewer files there are for the same dataset. Consequently, there is a trade-off between optimizing for object read performance, and the amount of parallelism achievable on a particular query. Large files are best for large scans as the query likely operates on sufficiently large number of files. For queries that are more selective and for which fewer files are operating, you may find that smaller files allow for more parallelism.
  • Data format. Redshift Spectrum supports various data formats. Columnar formats like Parquet can sometimes lead to substantial performance benefits by providing compression and more efficient I/O for certain workloads. Generally, format types like Parquet should be used for query workloads involving large scans, and high attribute selectivity. Again, there are trade-offs as formats like Parquet require more compute power to process than plaintext. For queries on smaller subsets of data, the I/O efficiency benefit of Parquet is diminished. At some point, Parquet may perform the same or slower than plaintext. Latency, compression rates, and the trade-off between user experience and cost should drive your decision.

To help illustrate how Redshift Spectrum performs on these large aggregation workloads, run a basic query that aggregates the entire ~3.7 billion record dataset on Redshift Spectrum, and compared that with running the query exclusively on Amazon Redshift:

SELECT uv.custKey, COUNT(uv.custKey)
FROM <your clickstream table> as uv
GROUP BY uv.custKey
ORDER BY uv.custKey ASC

For the Amazon Redshift test case, the clickstream data is loaded, and distributed evenly across all nodes (even distribution style) with optimal column compression encodings prescribed by the Amazon Redshift’s ANALYZE command.

The Redshift Spectrum test case uses a Parquet data format with each file containing all the data for a particular customer in a month. This results in files mostly in the range of 220-280 MB, and in effect, is the largest file size for this partitioning scheme. If you run tests with the other datasets provided, you see that this data format and size is optimal and out-performs others by ~60X. 

Performance differences will vary depending on the scenario. The important takeaway is to understand the testing strategy and the workload characteristics where Redshift Spectrum is likely to yield performance benefits. 

The following chart compares the query execution time for the two scenarios. The results indicate that you would have to pay for 12 X DC1.Large nodes to get performance comparable to using a small Amazon Redshift cluster that leverages Redshift Spectrum. 

Chart showing simple aggregation on ~3.7 billion records

So you’ve validated that Spectrum excels at performing large aggregations. Could you benefit by pushing more work down to Redshift Spectrum in your original query? It turns out that you can, by making the following modification:

The clickstream data is stored at a day-level granularity for each customer while your query rolls up the data to the month level per customer. In the earlier query that uses the day/month partition key, you optimized the query so that it only scans and retrieves the data required, but the day level data is still sent back to your Amazon Redshift cluster for joining and aggregation. The query shown here pushes aggregation work down to Redshift Spectrum as indicated by the query plan:

In this query, Redshift Spectrum aggregates the clickstream data to the month level before it is returned to the Amazon Redshift cluster and joined with the dimension tables. This query should complete in about 4 seconds, which is roughly twice as fast as only using the partition key. The speed increase is evident upon reviewing the SVL_S3QUERY_SUMMARY table:

  • Bytes scanned is 21.6X less because of the Parquet data format.
  • Only 90 records are returned back to the Amazon Redshift cluster as a result of the push-down, instead of ~66.2 million, leading to substantially less join overhead, and about 530 MB less data sent back to your cluster.
  • No adverse change in average parallelism.

Assessing the value of Amazon Redshift vs. Redshift Spectrum

At this point, you might be asking yourself, why would I ever not use Redshift Spectrum? Well, you still get additional value for your money by loading data into Amazon Redshift, and querying in Amazon Redshift vs. querying S3.

In fact, it turns out that the last version of our query runs even faster when executed exclusively in native Amazon Redshift, as shown in the following chart:

Chart comparing Amazon Redshift vs. Redshift Spectrum with pushdown aggregation over 3 months of data

As a general rule, queries that aren’t dominated by I/O and which involve multiple joins are better optimized in native Amazon Redshift. For instance, the performance difference between running the partition key query entirely in Amazon Redshift versus with Redshift Spectrum is twice as large as that that of the pushdown aggregation query, partly because the former case benefits more from better join performance.

Furthermore, the variability in latency in native Amazon Redshift is lower. For use cases where you have tight performance SLAs on queries, you may want to consider using Amazon Redshift exclusively to support those queries.

On the other hand, when you perform large scans, you could benefit from the best of both worlds: higher performance at lower cost. For instance, imagine that you wanted to enable your business analysts to interactively discover insights across a vast amount of historical data. In the example below, the pushdown aggregation query is modified to analyze seven years of data instead of three months:

SELECT c.c_name, c.c_mktsegment, t.prettyMonthYear, uv.totalRevenue
…
WHERE customer <= 3 and visitYearMonth >= 199201
… 
FROM dwdate WHERE d_yearmonthnum >= 199201) as t
…
ORDER BY c.c_name, c.c_mktsegment, uv.visitYearMonth ASC

This query requires scanning and aggregating nearly 1.9 billion records. As shown in the chart below, Redshift Spectrum substantially speeds up this query. A large Amazon Redshift cluster would have to be provisioned to support this use case. With the aid of Redshift Spectrum, you could use an existing small cluster, keep a single copy of your data in S3, and benefit from economical, durable storage while only paying for what you use via the pay per query pricing model.

Chart comparing Amazon Redshift vs. Redshift Spectrum with pushdown aggregation over 7 years of data

Summary

Redshift Spectrum lowers the time to value for deeper insights on customer data queries spanning the data lake and data warehouse. It can enable interactive analysis on datasets in cases that weren’t economically practical or technically feasible before.

There are cases where you can get the best of both worlds from Redshift Spectrum: higher performance at lower cost. However, there are still latency-sensitive use cases where you may want native Amazon Redshift performance. For more best practice tips, see the 10 Best Practices for Amazon Redshift post.

Please visit the Amazon Redshift Spectrum PoC Environment Github page. If you have questions or suggestions, please comment below.

 


Additional Reading

Learn more about how Amazon Redshift Spectrum extends data warehousing out to exabytes – no loading required.


About the Author

Dylan Tong is an Enterprise Solutions Architect at AWS. He works with customers to help drive their success on the AWS platform through thought leadership and guidance on designing well architected solutions. He has spent most of his career building on his expertise in data management and analytics by working for leaders and innovators in the space.

 

 

AWS Cost Explorer Update – Better Filtering & Grouping, Report Management, RI Reports

Post Syndicated from Jeff Barr original https://aws.amazon.com/blogs/aws/aws-cost-explorer-update-better-filtering-grouping-report-management-ri-reports/

Our customers use Cost Explorer to better understand and manage their AWS spending, making heavy use of the reporting, analytics, and visualization tools that it provides. We launched Cost Explorer in 2014 with a focus on simplicity – single click signup, preconfigured default views, and a clean user interface (take a look back at The New AWS Cost Explorer to see where we started). The Cost Explorer has been very popular and we’ve received a lot of great feedback from our customers.

Last week we launched a major upgrade to Cost Explorer. We’ve redesigned the user interface to optimize many common workflows including filtering, report management, selection of date ranges, and grouping of data. We have also included some default reports to make it easier for you to explore the costs related to your use of Reserved Instances.

Looking at Cost Explorer
Since pictures are reportedly worth 1000 words, let’s take a closer look! Cost Explorer is part of the Billing Dashboard so I can start there:

Here’s the Billing Dashboard. I click on Cost Explorer to move ahead:

I can open up Cost Explorer or access one of three preconfigured views. I’ll go for the first option:

The default report shows my EC2 costs and usage (running hours) for the past 3 months:

I can use the Group By menu to break the costs down by EC2 instance type:

I have many other grouping options:

The filtering options are now easier to access and to edit. Here’s the full set:

I can explore my EC2 costs in any set of desired regions:

I can filter and then group by instance type to see how my spending breaks down:

I can click on Download CSV and then process the data locally:

I can also exclude certain instance types from the report. Here’s how I exclude my m4.xlarge, t2.micro, and t2.nano usage:

Report Management
Cost Explorer allows me to customize my existing reports and to create new reports from scratch. I can click on Save As to save my customized report with a new name:

I can see and manage all of my reports on the Saved Reports page (The padlock denotes a default report that cannot be edited and then overwritten):

When I click on New report I can start from a template:

After I click on Create Report, I set up my date range and filters as desired, and click on Save As. I created a report that displays my year-to-date usage of several AWS database services (Amazon Redshift, DynamoDB Accelerator (DAX), Amazon Relational Database Service (RDS), and AWS Database Migration Service):

All of my reports are accessible from the Reports menu so I can check on my costs with a click:

We also simplified the process of selecting a range of dates for a report, including options to select common date ranges:

Reserved Instance Reports
Cost Explorer also includes a pair of reports that will help you to understand and optimize your usage of Reserved Instances. I don’t own an RI’s so I used screen shots supplied by the team.

The RI Utilization report allows you to see how much of your purchased RI capacity is being put to use (the dashed red line represents a utilization target that you can specify):

The RI Coverage report tells you how much of your EC2 usage is being handled by Reserved Instances (this time, the dashed red line represents the desired amount of coverage):

I hope you have enjoyed this tour of the updated Cost Explorer. It is available now and you can start using it today!

Jeff;

Vetter: Why Github can’t host the Linux Kernel Community

Post Syndicated from corbet original https://lwn.net/Articles/730184/rss

Daniel Vetter describes
how the kernel community scales
and why he feels that the GitHub model tends not to
work for the largest projects. “Unfortunately github doesn’t support
this workflow, at least not natively in the github UI. It can of course be
done with just plain git tooling, but then you’re back to patches on
mailing lists and pull requests over email, applied manually. In my opinion
that’s the single one reason why the kernel community cannot benefit from
moving to github. There’s also the minor issue of a few top maintainers
being extremely outspoken against github in general, but that’s a not
really a technical issue. And it’s not just the linux kernel, it’s all huge
projects on github in general which struggle with scaling, because github
doesn’t really give them the option to scale to multiple repositories,
while sticking to with a monotree.

Run Common Data Science Packages on Anaconda and Oozie with Amazon EMR

Post Syndicated from John Ohle original https://aws.amazon.com/blogs/big-data/run-common-data-science-packages-on-anaconda-and-oozie-with-amazon-emr/

In the world of data science, users must often sacrifice cluster set-up time to allow for complex usability scenarios. Amazon EMR allows data scientists to spin up complex cluster configurations easily, and to be up and running with complex queries in a matter of minutes.

Data scientists often use scheduling applications such as Oozie to run jobs overnight. However, Oozie can be difficult to configure when you are trying to use popular Python packages (such as “pandas,” “numpy,” and “statsmodels”), which are not included by default.

One such popular platform that contains these types of packages (and more) is Anaconda. This post focuses on setting up an Anaconda platform on EMR, with an intent to use its packages with Oozie. I describe how to run jobs using a popular open source scheduler like Oozie.

Walkthrough

For this post, you walk through the following tasks:

  • Create an EMR cluster.
  • Download Anaconda on your master node.
  • Configure Oozie.
  • Test the steps.

Create an EMR cluster

Spin up an Amazon EMR cluster using the console or the AWS CLI. Use the latest release, and include Apache Hadoop, Apache Spark, Apache Hive, and Oozie.

To create a three-node cluster in the us-east-1 region, issue an AWS CLI command such as the following. This command must be typed as one line, as shown below. It is shown here separated for readability purposes only.

aws emr create-cluster \ 
--release-label emr-5.7.0 \ 
 --name '<YOUR-CLUSTER-NAME>' \
 --applications Name=Hadoop Name=Oozie Name=Spark Name=Hive \ 
 --ec2-attributes '{"KeyName":"<YOUR-KEY-PAIR>","SubnetId":"<YOUR-SUBNET-ID>","EmrManagedSlaveSecurityGroup":"<YOUR-EMR-SLAVE-SECURITY-GROUP>","EmrManagedMasterSecurityGroup":"<YOUR-EMR-MASTER-SECURITY-GROUP>"}' \ 
 --use-default-roles \ 
 --instance-groups '[{"InstanceCount":1,"InstanceGroupType":"MASTER","InstanceType":"<YOUR-INSTANCE-TYPE>","Name":"Master - 1"},{"InstanceCount":<YOUR-CORE-INSTANCE-COUNT>,"InstanceGroupType":"CORE","InstanceType":"<YOUR-INSTANCE-TYPE>","Name":"Core - 2"}]'

One-line version for reference:

aws emr create-cluster --release-label emr-5.7.0 --name '<YOUR-CLUSTER-NAME>' --applications Name=Hadoop Name=Oozie Name=Spark Name=Hive --ec2-attributes '{"KeyName":"<YOUR-KEY-PAIR>","SubnetId":"<YOUR-SUBNET-ID>","EmrManagedSlaveSecurityGroup":"<YOUR-EMR-SLAVE-SECURITY-GROUP>","EmrManagedMasterSecurityGroup":"<YOUR-EMR-MASTER-SECURITY-GROUP>"}' --use-default-roles --instance-groups '[{"InstanceCount":1,"InstanceGroupType":"MASTER","InstanceType":"<YOUR-INSTANCE-TYPE>","Name":"Master - 1"},{"InstanceCount":<YOUR-CORE-INSTANCE-COUNT>,"InstanceGroupType":"CORE","InstanceType":"<YOUR-INSTANCE-TYPE>","Name":"Core - 2"}]'

Download Anaconda

SSH into your EMR master node instance and download the official Anaconda installer:

wget https://repo.continuum.io/archive/Anaconda2-4.4.0-Linux-x86_64.sh

At the time of publication, Anaconda 4.4 is the most current version available. For the download link location for the latest Python 2.7 version (Python 3.6 may encounter issues), see https://www.continuum.io/downloads.  Open the context (right-click) menu for the Python 2.7 download link, choose Copy Link Location, and use this value in the previous wget command.

This post used the Anaconda 4.4 installation. If you have a later version, it is shown in the downloaded filename:  “anaconda2-<version number>-Linux-x86_64.sh”.

Run this downloaded script and follow the on-screen installer prompts.

chmod u+x Anaconda2-4.4.0-Linux-x86_64.sh
./Anaconda2-4.4.0-Linux-x86_64.sh

For an installation directory, select somewhere with enough space on your cluster, such as “/mnt/anaconda/”.

The process should take approximately 1–2 minutes to install. When prompted if you “wish the installer to prepend the Anaconda2 install location”, select the default option of [no].

After you are done, export the PATH to include this new Anaconda installation:

export PATH=/mnt/anaconda/bin:$PATH

Zip up the Anaconda installation:

cd /mnt/anaconda/
zip -r anaconda.zip .

The zip process may take 4–5 minutes to complete.

(Optional) Upload this anaconda.zip file to your S3 bucket for easier inclusion into future EMR clusters. This removes the need to repeat the previous steps for future EMR clusters.

Configure Oozie

Next, you configure Oozie to use Pyspark and the Anaconda platform.

Get the location of your Oozie sharelibupdate folder. Issue the following command and take note of the “sharelibDirNew” value:

oozie admin -sharelibupdate

For this post, this value is “hdfs://ip-192-168-4-200.us-east-1.compute.internal:8020/user/oozie/share/lib/lib_20170616133136”.

Pass in the required Pyspark files into Oozies sharelibupdate location. The following files are required for Oozie to be able to run Pyspark commands:

  • pyspark.zip
  • py4j-0.10.4-src.zip

These are located on the EMR master instance in the location “/usr/lib/spark/python/lib/”, and must be put into the Oozie sharelib spark directory. This location is the value of the sharelibDirNew parameter value (shown above) with “/spark/” appended, that is, “hdfs://ip-192-168-4-200.us-east-1.compute.internal:8020/user/oozie/share/lib/lib_20170616133136/spark/”.

To do this, issue the following commands:

hdfs dfs -put /usr/lib/spark/python/lib/py4j-0.10.4-src.zip hdfs://ip-192-168-4-200.us-east-1.compute.internal:8020/user/oozie/share/lib/lib_20170616133136/spark/
hdfs dfs -put /usr/lib/spark/python/lib/pyspark.zip hdfs://ip-192-168-4-200.us-east-1.compute.internal:8020/user/oozie/share/lib/lib_20170616133136/spark/

After you’re done, Oozie can use Pyspark in its processes.

Pass the anaconda.zip file into HDFS as follows:

hdfs dfs -put /mnt/anaconda/anaconda.zip /tmp/myLocation/anaconda.zip

(Optional) Verify that it was transferred successfully with the following command:

hdfs dfs -ls /tmp/myLocation/

On your master node, execute the following command:

export PYSPARK_PYTHON=/mnt/anaconda/bin/python

Set the PYSPARK_PYTHON environment variable on the executor nodes. Put the following configurations in your “spark-opts” values in your Oozie workflow.xml file:

–conf spark.executorEnv.PYSPARK_PYTHON=./anaconda_remote/bin/python
–conf spark.yarn.appMasterEnv.PYSPARK_PYTHON=./anaconda_remote/bin/python

This is referenced from the Oozie job in the following line in your workflow.xml file, also included as part of your “spark-opts”:

--archives hdfs:///tmp/myLocation/anaconda.zip#anaconda_remote

Your Oozie workflow.xml file should now look something like the following:

<workflow-app name="spark-wf" xmlns="uri:oozie:workflow:0.5">
<start to="start_spark" />
<action name="start_spark">
    <spark xmlns="uri:oozie:spark-action:0.1">
        <job-tracker>${jobTracker}</job-tracker>
        <name-node>${nameNode}</name-node>
        <prepare>
            <delete path="/tmp/test/spark_oozie_test_out3"/>
        </prepare>
        <master>yarn-cluster</master>
        <mode>cluster</mode>
        <name>SparkJob</name>
        <class>clear</class>
        <jar>hdfs:///user/oozie/apps/myPysparkProgram.py</jar>
        <spark-opts>--queue default
            --conf spark.ui.view.acls=*
            --executor-memory 2G --num-executors 2 --executor-cores 2 --driver-memory 3g
            --conf spark.executorEnv.PYSPARK_PYTHON=./anaconda_remote/bin/python
            --conf spark.yarn.appMasterEnv.PYSPARK_PYTHON=./anaconda_remote/bin/python
            --archives hdfs:///tmp/myLocation/anaconda.zip#anaconda_remote
        </spark-opts>
    </spark>
    <ok to="end"/>
    <error to="kill"/>
</action>
        <kill name="kill">
                <message>Action failed, error message[${wf:errorMessage(wf:lastErrorNode())}]</message>
        </kill>
        <end name="end"/>
</workflow-app>

Test steps

To test this out, you can use the following job.properties and myPysparkProgram.py file, along with the following steps:

job.properties

masterNode ip-xxx-xxx-xxx-xxx.us-east-1.compute.internal
nameNode hdfs://${masterNode}:8020
jobTracker ${masterNode}:8032
master yarn
mode cluster
queueName default
oozie.libpath ${nameNode}/user/oozie/share/lib
oozie.use.system.libpath true
oozie.wf.application.path ${nameNode}/user/oozie/apps/

Note: You can get your master node IP address (denoted as “ip-xxx-xxx-xxx-xxx” here) from the value for the sharelibDirNew parameter noted earlier.

myPysparkProgram.py

from pyspark import SparkContext, SparkConf
import numpy
import sys

conf = SparkConf().setAppName('myPysparkProgram')
sc = SparkContext(conf=conf)

rdd = sc.textFile("/user/hadoop/input.txt")

x = numpy.sum([3,4,5]) #total = 12

rdd = rdd.map(lambda line: line + str(x))
rdd.saveAsTextFile("/user/hadoop/output")

Put the “myPysparkProgram.py” into the location mentioned between the “<jar>xxxxx</jar>” tags in your workflow.xml. In this example, the location is “hdfs:///user/oozie/apps/”. Use the following command to move the “myPysparkProgram.py” file to the correct location:

hdfs dfs -put myPysparkProgram.py /user/oozie/apps/

Put the above workflow.xml file into the “/user/oozie/apps/” location in hdfs:

hdfs dfs –put workflow.xml /user/oozie/apps/

Note: The job.properties file is run locally from the EMR master node.

Create a sample input.txt file with some data in it. For example:

input.txt

This is a sentence.
So is this. 
This is also a sentence.

Put this file into hdfs:

hdfs dfs -put input.txt /user/hadoop/

Execute the job in Oozie with the following command. This creates an Oozie job ID.

oozie job -config job.properties -run

You can check the Oozie job state with the command:

oozie job -info <Oozie job ID>

  1. When the job is successfully finished, the results are located at:
/user/hadoop/output/part-00000
/user/hadoop/output/part-00001

  1. Run the following commands to view the output:
hdfs dfs -cat /user/hadoop/output/part-00000
hdfs dfs -cat /user/hadoop/output/part-00001

The output will be:

This is a sentence. 12
So is this 12
This is also a sentence 12

Summary

The myPysparkProgram.py has successfully imported the numpy library from the Anaconda platform and has produced some output with it. If you tried to run this using standard Python, you’d encounter the following error:

Now when your Python job runs in Oozie, any imported packages that are implicitly imported by your Pyspark script are imported into your job within Oozie directly from the Anaconda platform. Simple!

If you have questions or suggestions, please leave a comment below.


Additional Reading

Learn how to use Apache Oozie workflows to automate Apache Spark jobs on Amazon EMR.

 


About the Author

John Ohle is an AWS BigData Cloud Support Engineer II for the BigData team in Dublin. He works to provide advice and solutions to our customers on their Big Data projects and workflows on AWS. In his spare time, he likes to play music, learn, develop tools and write documentation to further help others – both colleagues and customers alike.

 

 

 

New – GPU-Powered Streaming Instances for Amazon AppStream 2.0

Post Syndicated from Jeff Barr original https://aws.amazon.com/blogs/aws/new-gpu-powered-streaming-instances-for-amazon-appstream-2-0/

We launched Amazon AppStream 2.0 at re:Invent 2016. This application streaming service allows you to deliver Windows applications to a desktop browser.

AppStream 2.0 is fully managed and provides consistent, scalable performance by running applications on general purpose, compute optimized, and memory optimized streaming instances, with delivery via NICE DCV – a secure, high-fidelity streaming protocol. Our enterprise and public sector customers have started using AppStream 2.0 in place of legacy application streaming environments that are installed on-premises. They use AppStream 2.0 to deliver both commercial and line of business applications to a desktop browser. Our ISV customers are using AppStream 2.0 to move their applications to the cloud as-is, with no changes to their code. These customers focus on demos, workshops, and commercial SaaS subscriptions.

We are getting great feedback on AppStream 2.0 and have been adding new features very quickly (even by AWS standards). So far this year we have added an image builder, federated access via SAML 2.0, CloudWatch monitoring, Fleet Auto Scaling, Simple Network Setup, persistent storage for user files (backed by Amazon S3), support for VPC security groups, and built-in user management including web portals for users.

New GPU-Powered Streaming Instances
Many of our customers have told us that they want to use AppStream 2.0 to deliver specialized design, engineering, HPC, and media applications to their users. These applications are generally graphically intensive and are designed to run on expensive, high-end PCs in conjunction with a GPU (Graphics Processing Unit). Due to the hardware requirements of these applications, cost considerations have traditionally kept them out of situations where part-time or occasional access would otherwise make sense. Recently, another requirement has come to the forefront. These applications almost always need shared, read-write access to large amounts of sensitive data that is best stored, processed, and secured in the cloud. In order to meet the needs of these users and applications, we are launching two new types of streaming instances today:

Graphics Desktop – Based on the G2 instance type, Graphics Desktop instances are designed for desktop applications that use the CUDA, DirectX, or OpenGL for rendering. These instances are equipped with 15 GiB of memory and 8 vCPUs. You can select this instance family when you build an AppStream image or configure an AppStream fleet:

Graphics Pro – Based on the brand-new G3 instance type, Graphics Pro instances are designed for high-end, high-performance applications that can use the NVIDIA APIs and/or need access to large amounts of memory. These instances are available in three sizes, with 122 to 488 GiB of memory and 16 to 64 vCPUs. Again, you can select this instance family when you configure an AppStream fleet:

To learn more about how to launch, run, and scale a streaming application environment, read Scaling Your Desktop Application Streams with Amazon AppStream 2.0.

As I noted earlier, you can use either of these two instance types to build an AppStream image. This will allow you to test and fine tune your applications and to see the instances in action.

Streaming Instances in Action
We’ve been working with several customers during a private beta program for the new instance types. Here are a few stories (and some cool screen shots) to show you some of the applications that they are streaming via AppStream 2.0:

AVEVA is a world leading provider of engineering design and information management software solutions for the marine, power, plant, offshore and oil & gas industries. As part of their work on massive capital projects, their customers need to bring many groups of specialist engineers together to collaborate on the creation of digital assets. In order to support this requirement, AVEVA is building SaaS solutions that combine the streamed delivery of engineering applications with access to a scalable project data environment that is shared between engineers across the globe. The new instances will allow AVEVA to deliver their engineering design software in SaaS form while maximizing quality and performance. Here’s a screen shot of their Everything 3D app being streamed from AppStream:

Nissan, a Japanese multinational automobile manufacturer, trains its automotive specialists using 3D simulation software running on expensive graphics workstations. The training software, developed by The DiSti Corporation, allows its specialists to simulate maintenance processes by interacting with realistic 3D models of the vehicles they work on. AppStream 2.0’s new graphics capability now allows Nissan to deliver these training tools in real time, with up to date content, to a desktop browser running on low-cost commodity PCs. Their specialists can now interact with highly realistic renderings of a vehicle that allows them to train for and plan maintenance operations with higher efficiency.

Cornell University is an American private Ivy League and land-grant doctoral university located in Ithaca, New York. They deliver advanced 3D tools such as AutoDesk AutoCAD and Inventor to students and faculty to support their course work, teaching, and research. Until now, these tools could only be used on GPU-powered workstations in a lab or classroom. AppStream 2.0 allows them to deliver the applications to a web browser running on any desktop, where they run as if they were on a local workstation. Their users are no longer limited by available workstations in labs and classrooms, and can bring their own devices and have access to their course software. This increased flexibility also means that faculty members no longer need to take lab availability into account when they build course schedules. Here’s a copy of Autodesk Inventor Professional running on AppStream at Cornell:

Now Available
Both of the graphics streaming instance families are available in the US East (Northern Virginia), US West (Oregon), EU (Ireland), and Asia Pacific (Tokyo) Regions and you can start streaming from them today. Your applications must run in a Windows 2012 R2 environment, and can make use of DirectX, OpenGL, CUDA, OpenCL, and Vulkan.

With prices in the US East (Northern Virginia) Region starting at $0.50 per hour for Graphics Desktop instances and $2.05 per hour for Graphics Pro instances, you can now run your simulation, visualization, and HPC workloads in the AWS Cloud on an economical, pay-by-the-hour basis. You can also take advantage of fast, low-latency access to Amazon Elastic Compute Cloud (EC2), Amazon Simple Storage Service (S3), AWS Lambda, Amazon Redshift, and other AWS services to build processing workflows that handle pre- and post-processing of your data.

Jeff;