Hi Tharindu, Agreed .. lets plan to find some HW and do this when we can (will need at least 5 servers) .. I just want us not to be blocked on this.
--Srinath On Thu, Feb 9, 2012 at 2:01 PM, Tharindu Mathew <[email protected]> wrote: > Not testing for load at start lead to problems earlier. > > Testing for 20 clients and 10 M messages is not the same for testing for 500 > clients and 10 M messages. At this scale only, connection pooling, caching > and all of this matters. This is of course, many clients to one server, a > basic test that can be done easily. > > Anyway, we can post pone the testing to when we integrate BAM or CEP, but > please be aware of the implications. This was a situation where earlier BAM > was not scaling at all. > > > On Thu, Feb 9, 2012 at 1:43 PM, Amila Suriarachchi <[email protected]> wrote: >> >> >> >> On Thu, Feb 9, 2012 at 12:36 PM, Srinath Perera <[email protected]> wrote: >>> >>> Hi Suho, >>> >>> IMHO this covers about 90% of our usecases .. I think we should move on >> >> >> we do not required to do an performance study. Since anyway we don't have >> proper CEP or BAM service to consume events. >> >> But need to test for once client many server scenario to test for >> stability. >> >> thanks, >> Amila. >> >>> >>> >>> I do not mind doing a larger perf study later .. .. >>> >>> but getting the server finished from siddhi side and releasing it, CEP >>> perrf numbers, and integrating this with BAM are more urget IMHO >>> >>> --Srinath >>> >>> On Thu, Feb 9, 2012 at 12:29 PM, Suhothayan Sriskandarajah >>> <[email protected]> wrote: >>> > >>> > >>> > On Thu, Feb 9, 2012 at 12:00 PM, Amila Suriarachchi <[email protected]> >>> > wrote: >>> >> >>> >> >>> >> >>> >> On Thu, Feb 9, 2012 at 11:47 AM, Tharindu Mathew <[email protected]> >>> >> wrote: >>> >>> >>> >>> Suho, >>> >>> >>> >>> This does not measure the ability to handle load. Basically, what we >>> >>> did >>> >>> for load testing earlier was send around 2M messages, with >>> >>> concurrency as >>> >>> high increasing from 100 - 1500 clients. >>> >> >>> >> >>> >> How many back end servers and client programs used for that? >>> >> >>> >>> >>> >>> >>> >>> Your test is a measure of being able to consistently handle a stream >>> >>> of >>> >>> messages, which IMO, maybe important but is less interesting when >>> >>> handling >>> >>> load. >>> >> >>> >> >>> >> This test has measured the one client one server. yes it can be >>> >> improved >>> >> to one client to many servers and many clients to many servers. >>> >> >>> > >>> > This test also covers sending messages from many clients (2,5,10 & 20) >>> > to >>> > one server. >>> > Yes it could be improved to many clients to many server. >>> > If there are any specific requirements please suggest me a scenario, >>> > then I >>> > can do the testing and present the findings. >>> > >>> > Thanks >>> > Suho >>> > >>> >> But this test gives the through put this can handle one client to >>> >> server. >>> >> For one client scenario whether you add messages with one thread or >>> >> many >>> >> threads does not effect the performance if there are enough messages >>> >> to >>> >> send. >>> >> >>> >> thanks, >>> >> Amila. >>> >> >>> >>> >>> >>> To truly test the ability to handle loads we need to do a distributed >>> >>> load test with an extremely high number of clients and see the >>> >>> breaking >>> >>> point. Earlier, we could not find such a point with a single machine. >>> >>> >>> >>> >>> >>> On Thu, Feb 9, 2012 at 11:31 AM, Suhothayan Sriskandarajah >>> >>> <[email protected]> wrote: >>> >>>> >>> >>>> >>> >>>> >>> >>>> On Thu, Feb 9, 2012 at 11:24 AM, Amila Suriarachchi <[email protected]> >>> >>>> wrote: >>> >>>>> >>> >>>>> through put should not depend on the number of messages send. You >>> >>>>> need >>> >>>>> to send sufficient number of messages after server warmed up. >>> >>>>> >>> >>>>> Can you please first send around 2M messages and take the numbers >>> >>>>> for >>> >>>>> next 10M messages. >>> >>>>> >>> >>>>> And also try to test with number of back end servers. >>> >>>>> >>> >>>> >>> >>>> Yes, I'll work on this >>> >>>> >>> >>>> Thanks >>> >>>> Suho >>> >>>> >>> >>>>> >>> >>>>> thanks, >>> >>>>> Amila. >>> >>>>> >>> >>>>> >>> >>>>> On Wed, Feb 8, 2012 at 9:25 PM, Suhothayan Sriskandarajah >>> >>>>> <[email protected]> wrote: >>> >>>>>> >>> >>>>>> >>> >>>>>> According to the decisions made at the architecture mailing list >>> >>>>>> on >>> >>>>>> the subject "Serializing generic Siddhi events using Thrift", >>> >>>>>> I did some improvements to the code base of Agent component and >>> >>>>>> did a >>> >>>>>> performance testing. >>> >>>>>> >>> >>>>>> Here I did the tests by sending total of 1 million and 10 million >>> >>>>>> events using for loop, >>> >>>>>> in both cases I have sent events using 1,2,5,10 & 20 Clients. >>> >>>>>> >>> >>>>>> The results are as follows >>> >>>>>> >>> >>>>>> 1M 10M >>> >>>>>> 1 245118.1 269328.8 >>> >>>>>> 2 517509.1 739699.7 >>> >>>>>> 5 585823.1 923986.7 >>> >>>>>> 10 345582.3 418865.7 >>> >>>>>> 20 285008.6 368881.2 >>> >>>>>> >>> >>>>>> I have also attached the performance graph >>> >>>>>> >>> >>>>>> Regards >>> >>>>>> Suho >>> >>>>>> >>> >>>>>> >>> >>>>>> -- >>> >>>>>> S. Suhothayan >>> >>>>>> Software Engineer, >>> >>>>>> Data Technologies Team, >>> >>>>>> WSO2, Inc. http://wso2.com >>> >>>>>> lean.enterprise.middleware. >>> >>>>>> >>> >>>>>> email: [email protected] cell: (+94) 779 756 757 >>> >>>>>> blog: http://suhothayan.blogspot.com/ >>> >>>>>> twitter: http://twitter.com/suhothayan >>> >>>>>> linked-in: http://lk.linkedin.com/in/suhothayan >>> >>>>>> >>> >>>>>> >>> >>>>> >>> >>>>> >>> >>>>> >>> >>>>> -- >>> >>>>> Amila Suriarachchi >>> >>>>> >>> >>>>> Software Architect >>> >>>>> WSO2 Inc. ; http://wso2.com >>> >>>>> lean . enterprise . middleware >>> >>>>> >>> >>>>> phone : +94 71 3082805 >>> >>>>> >>> >>>> >>> >>>> >>> >>>> >>> >>>> -- >>> >>>> S. Suhothayan >>> >>>> Software Engineer, >>> >>>> Data Technologies Team, >>> >>>> WSO2, Inc. http://wso2.com >>> >>>> lean.enterprise.middleware. >>> >>>> >>> >>>> email: [email protected] cell: (+94) 779 756 757 >>> >>>> blog: http://suhothayan.blogspot.com/ >>> >>>> twitter: http://twitter.com/suhothayan >>> >>>> linked-in: http://lk.linkedin.com/in/suhothayan >>> >>>> >>> >>>> >>> >>> >>> >>> >>> >>> >>> >>> -- >>> >>> Regards, >>> >>> >>> >>> Tharindu >>> >>> >>> >>> blog: http://mackiemathew.com/ >>> >>> M: +94777759908 >>> >>> >>> >> >>> >> >>> >> >>> >> -- >>> >> Amila Suriarachchi >>> >> >>> >> Software Architect >>> >> WSO2 Inc. ; http://wso2.com >>> >> lean . enterprise . middleware >>> >> >>> >> phone : +94 71 3082805 >>> >> >>> > >>> > >>> > >>> > -- >>> > S. Suhothayan >>> > Software Engineer, >>> > Data Technologies Team, >>> > WSO2, Inc. http://wso2.com >>> > lean.enterprise.middleware. >>> > >>> > email: [email protected] cell: (+94) 779 756 757 >>> > blog: http://suhothayan.blogspot.com/ >>> > twitter: http://twitter.com/suhothayan >>> > linked-in: http://lk.linkedin.com/in/suhothayan >>> > >>> > >>> >>> >>> >>> -- >>> ============================ >>> Srinath Perera, Ph.D. >>> Senior Software Architect, WSO2 Inc. >>> Visiting Faculty, University of Moratuwa >>> Member, Apache Software Foundation >>> Research Scientist, Lanka Software Foundation >>> Blog: http://srinathsview.blogspot.com/ >>> Photos: http://www.flickr.com/photos/hemapani/ >>> Phone: 0772360902 >> >> >> >> >> -- >> Amila Suriarachchi >> >> Software Architect >> WSO2 Inc. ; http://wso2.com >> lean . enterprise . middleware >> >> phone : +94 71 3082805 >> > > > > -- > Regards, > > Tharindu > > blog: http://mackiemathew.com/ > M: +94777759908 > -- ============================ Srinath Perera, Ph.D. Senior Software Architect, WSO2 Inc. Visiting Faculty, University of Moratuwa Member, Apache Software Foundation Research Scientist, Lanka Software Foundation Blog: http://srinathsview.blogspot.com/ Photos: http://www.flickr.com/photos/hemapani/ Phone: 0772360902 _______________________________________________ Carbon-dev mailing list [email protected] http://mail.wso2.org/cgi-bin/mailman/listinfo/carbon-dev
