X

Learn how businesses like yours can begin to optimize for today and plan for tomorrow with Cloud-Ready IT Infrastructure

A Clear Path to Higher SAP Performance

Guest Author

Oracle Exadata X8M-2 demonstrates significant performance improvements in proof of concept

Overview

The Oracle Exadata Brazil Sales Team performed a very successful proof of concept (POC) with a large retailer running its SAP ECC and NetWeaver workloads on Oracle Exadata X8M and Private Cloud Appliance (PCA). The results demonstrate the combined capabilities of the Exadata X8M platform with PCA vs. Dell and HPE servers with HPE 3PAR All-Flash Storage: a 66% increase in SD (Sales and Distribution) transactions while running on 30% less infrastructure with a greater than 50% reduction in execution time.

Follow the Performance Leader

Customers have grown accustomed to performance leaps with every new generation of Oracle Exadata Database Machine due to improved processor speeds, architectural enhancements, and specialized algorithms in system software. With the introduction of the Exadata Database Machine X8M, performance gains are now orders-of-magnitude better than the previous generation largely due to the unique use of Intel® Optane™ Persistent Memory and a 100Gb RDMA over Converged Ethernet (RoCE) internal fabric. The result is, indeed, a leap in performance: 16 million OLTP read IOPS, <19 microseconds OLTP IO latency, and 560GB/sec analytic scan throughput.
 

The SAP customer journey with Exadata

How can this new Exadata generation help SAP customers? SAP ERP Central Component (ECC) systems often run mixed workloads of varying types. These differing workloads are where the Exadata X8M architecture leveraging persistent memory excels. This innovation is essential to keep batch chain jobs executing at consistently high performance, while maintaining user transactions at satisfactory response times for the business.

Let’s see how SAP workloads are handled by a typical do-it-yourself (DIY) system and Exadata as shown in Figure 1. In the DIY system, there are application, database servers, and external storage all connected via an Ethernet LAN. None of these components are optimized for Oracle Database and the LAN introduces delays. In contrast, Exadata is a full-stack, engineered system whose technology, including persistent memory and RoCE in the latest generation, are optimized and finely tuned to maximize Oracle Database performance.

How this works to enhance performance of SAP workloads: Exadata Storage servers transparently add persistent memory accelerators in front of flash memory. The Exadata Database servers uses Remote Direct Memory Access (RDMA) instead of I/O to read remote PMEM, which is automatically tiered and shared across databases using a cache for the hottest data.

Figure 1. Left: Do-it-yourself SAP environment. Right: Exadata X8M engineered system optimized with persistent memory to run all Oracle Database workloads significantly better.

With the use of Oracle Database 19c on Exadata X8M-2 (used in this PoC) we have the use of a third tier of information storage, in front of flash and behind the Database Nodes’ memory.  This is based on the new technology of Persistent Memory Data & Commit Accelerator that make use of Intel® Optane™ modules, associated with RDMA over Converged Ethernet (RoCE) 100GbE.

This technology also allows for lower log write latency, which is critical to OLTP performance, such as SAP applications. Faster log writes give way to faster transaction commit times and minimize any log write slowdown that may cause a commit backlog. The Automatic Commit Accelerator allows databases to perform “one-way” RDMA writes to PMEM on multiple Storage Servers, bypassing network and I/O software, interrupts and context switches. The performance benefits of the Exadata X8M are clearly demonstrated in the proof of concept (PoC) results below.

Clear Path with a PoC

Customers upgrading from legacy systems to an Exadata X8M often like to know in advance what performance gain they can expect. To address this need, the Oracle for SAP development team has run a series of benchmarks and published the results, in SAP Application Performance Standards (SAPS) values, in Oracle documentation and in SAPnotes. There are also special cases where a customer may request a PoC to help make the decision on which model and size of an Exadata they should invest in. In these cases, Oracle collaborates with the customer to design a test scenario involving a workload and real data to match the customer’s specific requirements.

The complete test scenario and results of the PoC are documented in a White Paper “SAP on Exadata X8M-2 - SAP ECC with Oracle Database on Exadata X8M-2 POC Results.

You can download the new white paper here:  https://www.oracle.com/a/ocom/docs/sap-on-exadata-x8m-2-white-paper.pdf

Note: When using Exadata X8M for SAP applications, it is extremely important that all SAP Bundle-Patches (SBP) and upgrade procedures cascading from SAP Note 2799900 for Oracle Database 19c local to the Exadata are also applied accordingly – which was the case for this PoC.

For more information on SAP on Oracle, please visit  https://www.oracle.com/solutions/sap/

For more Exadata-specific information https://www.oracle.com/engineered-systems/exadata/ or simply ask your Oracle Representative.

Authors:

Thomas Schuele is a Business Consultant and Program Manager at Oracle’s Oracle-for-SAP Competency Center in Walldorf, Germany. He has over 20+ years of experience with SAP where he has been managing and driving the continuous innovations from Oracle for SAP customers.

Brian Gruttadauria is responsible for Cloud Infrastructure as part of the Oracle Database & Autonomous Services Product Marketing group. Brian has 25+ years of experience running global development organizations. Prior to joining Oracle, he was responsible for the innovation and development of dozens of enterprise infrastructure and cloud products.

Be the first to comment

Comments ( 0 )
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.