SAP XI Performance Measurements

[Former Member]

Hi All,

In the absence of publicly available SAP XI Performance benchmarks, I conducted a few performance measurements in our labs.

Here is the result:

Test case consists of reading from a Database through the DB Adapter, getting into XI Server, conduct mapping / transformation, and write it to the File Adapter.

The Mapping Complexity is low (simple assignments, string concatenations).

- Test Case 1:

Read 100,000 rows from Database into XI server, conducted mapping & transformation, and wrote to the File Adapter.

Target Document Created: 25 MB

Total Time taken: 86 Seconds

Time taken in Mapping: 80 Seconds

- Test Case 2:

Read 40,000 rows from Database, map/transform, and write to FileAdapter.

Target Document Created: 10 MB

Total Time taken: 17 Seconds

Time taken in Mapping: 14 Seconds

Platform: Central Instance on a dual CPU Intel Xeon (2.4GHz), 8GB RAM, Windows Adv Server. Database (SQLServer) on the same host.

Conclusions:

- XI Server performance is "reasonably good" considering it handled a 25MB document in 86 seconds on a mid range server class machine.

- Mapping is non-linear. Need to further investigate with more realistic and complex mapping scenarios (string operations, loops, node counts, value-mappings, lookups etc.)

Hope this information proves useful in making customer presentations and a case for XI.

thanks,

Manish

[Former Member]

Hello,

I was just wondering what is the configuration of XI, you are using right now: what is the hardware used, how XI IS and XI AF are installed, and are there any specification configuration you made to J2EE?

Thank you in advance,

Dmitriy Mindich

[OliverDeckert]

Hi Manish,

looking simply at processing time might be unsufficient.

In this testcase, as mapping is the key in processing time, you should also mention garbage collection behaviour of the scenario.

Probably, when looking at the document size, memory usage on java level plays a role and might influence mapping runtime in a non-linear way.

Regards, Oliver

[Former Member]

Hi Oliver

That is correct, we cannot conduct performance measurements in isolation.

My test is just a single one-off test where I measured the processing time for 1 transaction in isolation.

For tests to be meaningful in the context of XI Server performance, reliability and availability, we need to see the effects on the JVM memory usage, garbage collection and effects on the server availability.

I have seen applications where the memory usage grows linearly with a low constant transaction load, ultimately leading to an OutOfMemory error. This is a candidate for a 24x7 test measuring availability.

Another application behaviour under bursty load conditions is that the memory usage increases as a spike. The garbage collector kicks in and the memory usage comes down to a lower level. However, all objects are not released, and gradually with time, we end up with an OutOfMemory error.

We also need to test server availability under bursty load conditions - server may stop accepting messages temporarily when swamped with a temporary high load. A well designed application should accept messages under heavy load, release memory after servicing all requests, and not lose any requests (measure of reliability).

Hopefully, SAP is conducting tests on these fronts and would share them with customers and partners.

thanks,

Manish

[Former Member]

Hi,

That is wonderful, Manish.

Thanks a lot for the information. Please keep sharing your findings so that others can make use of it.

Warm regards,

Venki.