Post Syndicated from Martin Yip original https://aws.amazon.com/blogs/compute/estimating-amazon-ec2-instance-needed-when-migrating-erp-from-ibm-power/
This post courtesy of CK Tan, AWS, Enterprise Migration Architect – APAC
Today there are many enterprise customers who are keen on migrating their mission critical Enterprise Resource Planning (ERP) applications such as Oracle E-Business Suite (Oracle EBS) or SAP running on IBM Power System from on-premises to the AWS cloud platform. They realize running critical ERP applications on AWS will enable their business to be more agile, cost-effective, and secure than on premises. AWS provides cloud native services that streamline your ability to adopt emerging technologies, enabling greater innovation, and faster time-to-value.
However, it is essential to note that both IBM Power System and Amazon Elastic Cloud Compute (EC2) are using different CPU processor architectures. Amazon EC2 instances are running on either x86 or ARM-based processors while IBM Power System is running on IBM Power processors. As a result, there are no direct performance benchmarks mapping the two platforms, or sizing for enterprise customers who intend to migrate their application running on IBM Power System to Amazon EC2.
The purpose of this blog is to share a methodology along with example of how to estimate the sizing of mission critical applications running on IBM Power System to Amazon EC2 running on x86 platform.
In order to normalize the CPU performance between IBM Power and Intel x86 processors, we may refer to SAP Application Performance Standard (SAPS) for performance benchmark purposes. SAPS is a hardware-independent unit of measurement that describes the performance of a system configuration in the SAP environment. It is derived from the Sales and Distribution (SD) two-tier benchmark, where 100 SAPS is defined as 2,000 fully business processed order line items per hour.
In technical terms, this throughput is achieved by processing 6,000 dialog steps (screen changes), 2,000 postings per hour in the SD Benchmark, or 2,400 SAP transactions.
In the SD benchmark, fully business processed means the full business process of an order line item: creating the order, creating a delivery note for the order, displaying the order, changing the delivery, posting a goods issue, listing orders, and creating an invoice.
In this methodology, we compare the performance sizing by using SAPS which is not released by other ERP software such as Oracle EBS. However, we believe that this methodology will be most pertinent of indirect performance benchmark from an ERP software perspective as most of the enterprise businesses have similar process of ordering.
Discussion by Example
We will walk you through a real example to demonstrate how you can translate and migrate an SAP ERP 6.0 or Oracle EBS application database that is running on IBM Power 795 (IBM P795) with IBM AIX from on-premises to AWS cloud environment by choosing the right size of Amazon EC2 instances.
Step 1: IBM P795 System Performance Analysis
Users can leverage IBM nmon Analyzer which is a free tool to produce AIX performance analysis reports running on IBM Power System. The tool is designed to work with the latest version of nmon, but it is also tested with older versions for backwards compatibility. The tool is updated whenever nmon is updated, and at irregular intervals for new function. It allows users to:
- View the data in spreadsheet form
- Eliminate “bad” data
- Automatically produce graphs for presentation to clients
Below is the performance analysis being captured for CPU, memory, disk throughput, disk Input / Output (I/O), and network I/O during the system’s database peak demand at month end closing – using IBM nmon Analyzer.
Figure 1: Actual Number of Pyhsical Core of CPU Utilization Analysis on IBM P795
Figure 2: Memory Usage Analysis on IBM P795
Figure 3: Total Disk Throughput Analysis on IBM P795
Figure 4: Total Disk I/O on IBM P795
Figure 5: Network I/O Analysis on IBM P795
Step 2: CPU Normalization
Users can find the SAPS results for different hardware manufacturers from this link. Table 1 below shows how to calculate the processing power for single CPU core with respect to different hardware manufacturers. We have chosen Amazon EC2 instances r5.metal and r4.16xlarge as potential candidates because they provide memory sizing which best fits the current demand of 408 GB – refer to Figure 2.
|Instance Type||SAPS (2- Tier)||Memory (GB)||Total CPU Cores||SAPS/Core|
Table 1: Calculate the SAPS Processing Power for Single Core of Physical CPU
In order to calculate the Normalization Factor for any targeted Amazon EC2 instance with respect to source system of IBM Power system, we will apply Equation 1 as shown below:
Normalization Factor = [SAPS Per Core] Amazon EC2 / [SAPS Per Core] IBM POWER System (1)
Therefore, by inserting the SAPS/Core being calculated from Table 1 into Equation (1):
- The normalization factor for 1 core of Amazon EC2 of r5.metal to IBM P795 is equivalent to 1.08
- The normalization factor for 1 core of Amazon EC2 of r4.16xlarge to IBM P795 is equivalent to 0.79
Table 2 below shows how to calculate normalized total number of CPU cores being allocated to each system. We need to make sure that the normalized total number of CPU cores from selected Amazon EC2 instance must be greater than or equal to the assigned entitlement capacity of IBM P795 in this case.
|Instance Type||Normalization Factor (A)||Total # CPU Cores (B)||Normalized Total # CPU Core (A X B)|
Table 2: Normalized Total Number of CPU Core
Figure 1 indicates that the peak to meet the total processing power of database ERP running on IBM P795 is 29 cores of physical CPU as compared to assigned entitlement capacity of 32. As a result, the Amazon EC2 instance, r5.metal, should be the right candidate which will deliver sufficient processing power with enough buffer for future business growth in next 12 to 24 months.
Step 3: Amazon EC2 Right Sizing
Next, we need to check and ensure that other specifications such as network I/O, disk throughput and disk I/O will meet the current peak demand of the application. Table 3 shows the different specifications of r5.metal . It is clear from Table 3 that r5.metal meets all the demand requirements of the database ERP which is running on IBM P795.
|Specifications||Amazon EC2 Instance Type of r5.metal|
|Physical Core of CPU||48 > 29 (current CPU utilization)|
|RAM (GiB)||768 > 408 (current memory usage)|
|Network Perf (GBps)||3 > 0.3 (current network I/O throughput)|
|Dedicated EBS BW (GBps)||~ 2.4 > 2.0 (current disk throughput)|
|Maximum IOPS (16 KiB I/O)||80,000 > 25,000 (current disk IO consumption)|
Table 3: EC2 r5.metal Specifications
The methodology and example discussed in this blog gives a pragmatic estimation for optimal performance by selecting the right type and size of Amazon EC2 instances when you plan to migrate from IBM Power System, which is running mission critical ERP application, from on-premises datacenter to AWS cloud. This methodology has been applied to many enterprise cloud transformation projects and delivered more predictable performance with significant TCO savings. Additionally, this methodology can be adopted for capacity planning and helps enterprises establish strong business justifications for heterogeneous platform migration.