Benchmark Fix8

[Ludo]

Hello,

I open a new thread for follow the performance on the fix engine fix8.

I followed the test/performance page http://fix8.org/performance.html

I have test on : 

FIX8 0.6.5

gcc version 4.4.5

On VPS with shared memory 512Mb

Debian - Linux VM1 2.6.32-5 x86_64 GNU/Linux

16 CPUs CPU E5-2640 0 @ 2.50GHz (4988.0 bogomips) so total 79808 bogomips maybe shared but i don't know the %...

I have preload 50000 msgs ... for more messages i need more memory ;)

0 NewOrderSingle msgs currently preloaded.

Enter number of NewOrderSingle msgs to preload:50000

50000 NewOrderSingle msgs preloaded.

50000 NewOrderSingle msgs sent

0 NewOrderSingle msgs remaining.

88000 ExecutionReport msgs received

And the resultat :

av. decode(server): 0.00003573 sec/msg, 1.79 secs, 50011 msgs -----> 35.5 us/msg 3.35

av. encode(server): 0.00024278 sec/msg, 21.39 secs, 88087 msgs ----> 242 us/msg

av. decode(client): 0.00002724 sec/msg, 2.40 secs, 88077 msgs -------> 27.2 us/msg

av. encode(client): 0.00002653 sec/msg, 1.33 secs, 50011 msgs --------> 26.5 us/msg 

Strange resultat for my encode server ?!

anyway good taff for fix8 team :)

Regards

Ludo.

[dakka]

Hi,
50000 is too small a sample. There will be a lot of jitter which that small sample will not eliminate. You need to perform at least 500000 (that's 5 x 10^5).

If you are using a VPS you will not get max burst CPU performance. You will also get a lot of cache line pollution resulting in much worse performance than you might expect with a machine that has the reported specs you list.

I don't imagine a production environment would use VPS. See if you can't get you hands on something more suitable for this sort of test.

\dakka