It would be an understatement to say server infrastructure choices are richer than ever and increasing all the time, even in the often staid and conservative world of SAP implementations. The question that keeps many people up at night is how to maximize their investments in SAP infrastructure in a way that fits within their broader IT strategy. Choosing the best-performing SAP server platform can provide peace of mind.
Like most things, the answer depends on what investment they want to maximize. People tend to apply the easiest frame of reference to things, and when considering servers, they often use server performance as the benchmark. While server performance can be as useful as comparing the horsepower of different engines, there are many deficiencies to such an analysis. In this article, I'll outline some of the other measures that can affect the purchase decision.
Understanding CPU options and server performance
The SAP landscape includes numerous server manufacturers, all of whom base their servers on four chip architectures: x86 (32bit/64bit), Power, SPARC and Itanium.
These four architectures (a fifth, PA-RISC, merits an honorable mention) provide the customer with a choice -- in the beginning -- but each has its own history, pluses and minuses. There are powerful lock-ins associated with each architecture, and so decisions must be made with caution.
While there are many benchmarks associated with server performance, SAP created the SAP Application Performance Standard (SAPS) rating, a hardware-independent unit of measure that describes the ability of a system to complete a fixed amount of work. And it's not the only benchmark created by SAP; there are benchmarks for power consumption, data-load performance and front-end performance. These sadly underutilized benchmarks could be increasingly important, as they reflect powerful ways to look at server infrastructure and total cost of ownership (TCO).
The table below shows the current top SAPS ratings for each processor architecture, as released by SAP:
(Source: SAP SAPS benchmarks)
It's difficult to determine from the table which server has the best performance because the configurations of the servers are wildly different. However, it is possible to get a clearer picture by downloading the SAPS ratings, importing them into a spreadsheet and doing some filtering.
For example, the table below shows the top five benchmarked server architectures after applying two filters (12,000-13,000 users, and version EhP5 of SAP's ECC 6 ERP software):
Now, after filtering the above fields (more fields are available in the original data set), there is consistency in server configurations and SAPS ratings.
But note: The above benchmarks are not applicable to systems that run the SAP HANA in-memory technology. The benchmarks are based on ECC6 EhP5, which is not a release supported for Business Suite on HANA. Furthermore, SAP doesn't use the standard SAPS benchmark for HANA systems, and instead uses one called Enhanced Mixed Load (BW-EML).
More on SAP servers
Read a handbook on SAP server virtualization
Understand SAP system performance
See a tip on SAP disaster recovery
The BW-EML benchmarks emphasize three capabilities: near real-time reporting, ad-hoc reporting, and lower TCO through reduced data redundancy and required components.
As SAP and hardware providers start to increase the presence of Business Suite on HANA, this benchmark will be extended or replaced to reflect online transaction processing and online analytical processing workloads on the same server.
Look beyond the benchmarks to OS, database support
The above table also reveals that the top five systems in SAPS performance are very similar in terms of specification. This means other factors have to be taken into consideration.
Two major ones are vendor lock-in and the choice of operating systems that are available on each architecture to support different applications. Market share data provides some guidance on these.
The following table shows the market shares of x86 and RISC processors. Over the past five years, x86 processors have been increasing in market share for a number of reasons, one being that cloud service providers have been major purchasers of commodity x86 hardware. In addition, x86 vendors Intel and AMD have made significant strides in matching the performance of the most expensive RISC competitors, which has increased the use of commodity x86 hardware.
(Sources: Intel, Gartner)
Here's the breakdown of the RISC market shares:
(Sources: ITCandid, Gartner, Oracle)
Another reason for the increase in the x86 market can be seen in the final table, which shows the flexibility of each architecture by factoring in their support for different operating systems and databases. This table, compiled from the SAP Product Availability Matrix for the NetWeaver 7.x versions, shows that x86 is the most flexible architecture. Still, it's important to be cautious about jumping to conclusions, as certain databases are optimized for certain operating systems. For example, IBM DB2 is optimized for AIX. Similarly, Oracle Database is optimized for Solaris, and HANA is optimized for Linux on x86. The choice of database should be as close to optimal as possible, despite vendor financial incentives that may suggest otherwise.
(Source: SAP Product Availability Matrix for NetWeaver 7.x)
Other factors that affect server performance have not been touched on in this piece, including the penalty incurred from server virtualization, and external factors such as user interfaces and geographic distribution of users. Hopefully this short guide has provided useful insight into some of the considerations that must be taken into account when evaluating a server platform's performance.
About the author:
Chris Kernaghan is a senior SAP technical architect at CapGemini, the global consulting, technology and outsourcing firm.