Re: [Quickfix-developers] Performance of QuickFix/J vs QuickFix (C++)

[Robert P. <rob...@gm...>]

Also, for these number to be better interpreted, you'll need to be
more specific about the number of CPUs, and settings and version of
the JVM.  Also, what kind of logging are you using? What platform?

Otherwise it's hard to tell if you are testing QuickFIX or are seeing
effects of garbage collecting and such in the JVM itself, or perhaps
of File IO due to logging and persistence.

The very disparate values in the results, especially latency, say to
me that you are hitting system performance bottlenecks, and not really
testing QuickFIX/J at this point.


On 2/4/07, Robert Parrott <rob...@gm...> wrote:
> Hi Vikram, can you provide some units here? Are these microseconds?
>
> On 2/1/07, Vikram Venkatesan <rer...@ya...> wrote:
> > QuickFIX Documentation: http://www.quickfixengine.org/quickfix/doc/html/index.html
> > QuickFIX Support: http://www.quickfixengine.org/services.html
> >
> >
> > Hi,
> >     I have done some performance tests for Quickfix/J 1.0.4. and got some
> > performance results. Please let me know whether the performance results are
> > acceptable, and are expected.
> >
> > Started Client1 (Trader) and Client2(CounterParty) on the same machine
> > (Architecture : i386 / Memory : 4 GB ).
> >
> > 1.  Throughput testing
> >     Description :  Placed 100,000 orders from Trader and got all the orders
> > filled by the Counterparty (i.e. received all execution reports). The total
> > time taken is noted, and the throughout is calculated(orders per second) for
> > 100,000 orders.
> >
> > Average throughput : 3600 transactions / second.
> > (where transaction = sending order and receiving it's execution report).
> >
> > 2. Ping pong mode of testing
> >    Description : Placed 'N' orders from Trader. On receiving execution
> > report for all the 'N' orders, the next 'N' orders are placed and so
> > on...sort of BLOCKING order placement.
> > Latency :
> >           On receiving the execution report for the 'N' th order, the time
> > difference between the current time and the PLACING TIME of that order is
> > calculated and this is taken as the LATENCY for an order.
> >
> > Throuthput calculation:
> >       The total time taken for 'N' orders to be filled is taken down. From
> > this the throughput is calculated.
> >
> > Batch size = The value of 'N'.
> >
> > # orders  Batch size  Min Latency  Max Latency  Avg latency Min throughput
> > Max throughput  Avg throughput
> > 100000       1               1          1009              504             2
> > 1000                  575
> > 100000       2               1          1028              501            22
> > 2000                  1728
> > 100000       5               1          1009              501            90
> > 5000                  3999
> > 100000       10             1          1028             499               62
> > 10000                4300
> > 100000       100           9          1042             518               200
> > 9090                 5126
> > 100000       1000          28        1038             529               3058
> > 6802                 5778
> >
> >
> > - Vikram
> > --
> > View this message in context: http://www.nabble.com/Performance-of-QuickFix-J-vs-QuickFix-%28C%2B%2B%29-tf3134559.html#a8746000
> > Sent from the QuickFIX - Dev mailing list archive at Nabble.com.
> >
> >
> > -------------------------------------------------------------------------
> > Using Tomcat but need to do more? Need to support web services, security?
> > Get stuff done quickly with pre-integrated technology to make your job easier.
> > Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
> > http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
> > _______________________________________________
> > Quickfix-developers mailing list
> > Qui...@li...
> > https://lists.sourceforge.net/lists/listinfo/quickfix-developers
> >
>