As SAP applications grow more sophisticated, flexible, and integrated, the hardware resources required to support them have evolved as well. To keep their software operating optimally, companies must ensure that the right hardware is in place early on. That’s why determining the sizing requirements of your software before you purchase new hardware is so critical.
In the past, companies tended to look at throughput KPIs to determine which hardware resources would be right to support their software installations. However, there is now a new KPI that companies must also consider: single computing unit (SCU) performance. This refers to the processing power of a single computing unit within a system. These SCUs could include a single thread within a multi-threaded CPU, a core within a multicore CPU, or any other individual unit that makes up a CPU. Hardware that has good SCU performance helps ensure faster business process response times.
To help customers make informed hardware purchase decisions, SAP has been working together with its hardware partners to improve awareness of the importance of SCU performance and to make certain that customers recognize which SAP solutions benefit most highly from hardware that has good SCU performance. To that end, SAP has added SCU performance recommendations into our Quick Sizer tool (see sidebar below).1
SCU Performance KPI in Quick Sizer
From Quick Sizer version 24 (which went live on May 31, 2011) on, SAP has added a new sizing KPI for SCU performance. Now, each time customers conduct a sizing in Quick Sizer, they will see in the result view a new column that contains SCU performance class recommendations (see Figure 1). In close cooperation with our technology partners, SAP defined the three classes accordingly:
- Class A means that the SAP solution benefits from good SCU performance
- Class AA means that the SAP solution benefits from very good SCU performance
- Class AAA means that the SAP solution benefits from excellent SCU performance
||Quick Sizer results include SCU performance class.
For example, Figure 1 shows that a customer would need a total of 9,200 SAPS2 to run SAP CRM and should run the system on hardware that falls into SCU performance class AAA.
Once customers are aware that a certain application would benefit from hardware that has good SCU performance, the next challenge is to find the right hardware (see sidebar). A good place to start is to check the SAP Standard Application Benchmarks (www.sap.com/benchmark).3 These benchmarks run on partner hardware configurations to certify what these configurations can handle. However, these benchmarks define only the throughput power of the total system without explicitly measuring the SCU performance.
That’s why, as with any results from Quick Sizer, it is also very important that customers discuss their SCU performance requirements with their hardware vendors to ensure that the SCU performance of the hardware is sufficient to fulfill response time expectations. You can find more information about SCU performance and how hardware vendors can be contacted in SAP note 1501701.
It’s also important to keep in mind how hardware considerations, like SCU performance, clock rates, throughput, and other metrics, are continuing to evolve.
Evolving System Performance Considerations
Nowadays, the fastest SCUs have a capacity of roughly 2,000 SAPS. However, for some years now, the performance of an individual SCU has remained stable or has increased only very slightly (see Figure 2). At this point, it seems unlikely that individual SCU performances will become significantly faster in the foreseeable future.
||The development of single computing unit performance from 1996 to 2010 (Source: www.sap.com/benchmark).
In addition, as you can see in Figure 3, system clock rates have decreased since 2005. In the past it was possible to significantly increase SCU performance by increasing the processor clock speed; now, increasing clock speed has become much more difficult and that workaround is no longer possible. Instead, hardware vendors are investigating and investing in other areas, such as the introduction of multicore architectures.
These factors mean that future improvements to system performance will likely depend on other factors rather than on a continued performance improvement in individual SCUs. For instance, software developers will be challenged to write software that scales with the number of cores in a system and ensure that their solutions are designed in a way that the speed of hardware will not become a limiting factor.
In addition, for several years, a computer system’s ability to handle more and more throughput has continued to grow (mainly due to the introduction of parallelization and multicore technologies). We compared internal and published benchmark results from 1996 to 2011 and found that the total number of overall SAPS is increasing (see Figure 4).4
||The development of the performance of complete computer systems between 1996 and 2011 (Source: www.sap.com/benchmark).
The general trend is that developers will continue to design computer systems with more and more cores and threads per CPU. So while individual SCU performance is not likely to increase, the overall number of SCUs will. However, it is important to remember that adding these cores and threads might not speed up the existing SAP software in a significant way, depending on the SAP solution that is being used.
In addition to its throughput KPI, SAP has now introduced the SCU performance classes, which are crucial metrics for those who are looking for the right hardware to optimally support their software and solutions. I recommend that you explore the Quick Sizer tool, which evaluates both KPIs, and become familiar with its functionality before your next hardware implementation.
For more information about Quick Sizer, visit the SAP Service Marketplace at http://service.sap.com/quicksizing.
Sebastian Schmitt (firstname.lastname@example.org) joined SAP in 2007 after studying at the Universities of Cologne, Cáceres, and Barcelona. He has been a member of the Performance and Scalability team since 2008, and is responsible for hardware sizing from a product management perspective and for the cooperation of SAP and hardware vendors in the area of sizing.
1. See “Spending Too Much on Hardware?” by Sebastian Schmitt and Dagmar Kirsamer in the October-December 2009 issue of SAPinsider. [back]
2. SAP Application Performance Standard (SAPS) is a hardware-independent unit of measurement that describes the performance of a system configuration in the SAP environment. More information is available at www.sap.com/benchmark. [back]
3. See “Take the Guesswork Out of Your Next Hardware Purchase: Depend on Experienced, Impartial SAP Standard Application Benchmarks to Find the Technology that Best Fits Your Business Needs” by Dr. Ulrich Marquard in the April-June 2008 issue of SAPinsider. [back]
4. The Sales & Distribution (SD) Benchmark was used to compare the throughput of these different systems. More information can be found at www.sap.com/benchmark. [back]