Difference between revisions of "DPL:Load Test"

Revision as of 07:23, 7 October 2008

Various Load Tests

Please use with care as we do not want to put extreme load to the web server.

• We have 100 documents named LoadTest_001 .. LoadTest_099.
• Each document has 20 chapters
• each chapter has 40 text lines of 72 characters

So, a document is around 60 kBytes, the whole text volume is around 6 MBytes

Each document makes 21 template calls (one per chapter and one at the end).

As this is a shared server results will not be the same all the time.

Test A

We extract 3 chapters from every document; This takes around 10 seconds typically but I have also observed response times below 5 seconds. The html output is 900 kBytes, so you need a fast connection, otherwise you will measure the time to get the result over the wire instead of the time to generate it.

Execute: DPL:Load Test A

Test B

We just pick up the template calls. This result is around 200 kBytes. Note that we process 2100 template invocations. Response time is usually in the same order as above, maybe a little faster. You may observe that the second executiuon of the test is considerably faster than the first one. Probably this has to do with MySQL caching strategies.

Note: actually we restricted the test to 10% of the load described above. Due to some configuration limits of this webserver it may happen otherwise that the operating system throws the process out.

Execute: DPL:Load Test B

Test C

We just produce the list of LoadTest articles without steppping into them for analysis. This is fast (2-3 seconds) although we use a LIKE expression. This may change if you have more documents in your wiki.

Execute: DPL:Load Test C

Test D

We select the same articles based on the fact that they all use the same template. This requires some database join. There is no big difference to the test above.

Execute: DPL:Load Test D