Tigase 5
Tigase 5.0.0 Beta released. Finally, all the automatic tests passed, the code has been working on a few live installations for some time already. It looks like the new version is Beta ready.
71, kobit//tigase.org, 2010-03-09 01:06
Functional tests completed
Functional tests completed... All functional tests for 3 databases and all authentication connectors passed successfully. Results are being copied to the website now. One step closer for the new release.
70, kobit//tigase.org, 2010-03-08 15:17
Tigase 5
Tigase 5.0.0 release delayed. I am sorry but it looks like the release will be delayed for about 2 weeks. There are 2 bugs discovered which have to be fixed before the release. I will try to publish kind of Beta earlier for people who want to use the new API available to develop own code.
69, kobit//tigase.org, 2010-02-14 21:19
I am not familiar with them
I am not familiar with them but after a brief reading I think they are actually dedicated to website performance tests which is very different from XMPP services testing.
Is softwaretestqa.com the
Is softwaretestqa.com the best place to find performance testing help?
We were able to analyse the
We were able to analyse the test report and here is what we had identified:
Ran a test for 2000 users from 2 client machines using distributed load technique.1976 users were able to get connected successfully, but as soon as the users started authenticating (tr_authenticate) connections started dropping and the number of connections that dropped initially were 859 and stayed the same till the completion of the test.
We are trying to figure out why these connections dropped during this transaction(tr_authenticate)? & also why all the users couldnt get connected successfully.
Can you please help us out with this?
If needed i would send across the log files of the server collected during this test,the reports generated for this test and any other details that are needed.
Thanks in advance.
SR
Once again thank you very
Once again thank you very much for the very helpful information.We have also gone through the Tsung user manual and its documentation.
As for the initial test machine configuration We are using two dual core machines each with 2.0GB RAM and 1.86GHz proc.Strange thing is that we are trying to run the tests for 1000 users(test for connect,Authenticate & presence:initial only) on each test machine using distributed load balancing as given in Tsung manual, but we are unable to test successfully for more than ~1036 in almost every time we had run the test.Error in most cases 'error_unknown'.
Since we are running the tests on Linux machine we have also configured the machine to run tests for high load as per your suggestion, which seems to have resolved the error "Too many Open Files..."
Many Thanks,
SR
The number of the load the
The number of the load the server can handle depends on a few factors. The most important probably is the roster size as it implies the memory consumption used to cache the user roster and CPU usage for presence packets processing.
ad. 1. There are a few articles about Tigase monitoring on the website: 500k user connections or Setting up remote monitoring in the server or Tigase Server monitoring next step or Tigase server monitoring. For the purpose of the load tests the best option is to combine jconsole monitoring which gives you insight in the resources usage like RAM or CPU with server internal statistics which gives you more details how the server works and what are the bottlenecks.
ad. 2. If you use tsung for your tests you should refer to it's documentation for all the settings regarding the number of connections and the load you can generate. Tsung is not very efficient in resources usage. You can expect to generate max 50k connections on machine with 2-4GB of RAM. Depending on the load you want to generate you also need at least 2 cores per 50k connections. Of course there is also a TCP/IP limit on the number of client ports you can open per single IP address but if you have a powerful client machine you can setup it with multiple IP addresses. In practice, you generally need at least 1 machine for generating client load per each 100k client connections.
Regarding the error from the log files it is what it says it is: Too many open files. You have to prepare the operating system configuration for high load. If you use a Unix based system like Linux you can use this guide which described the basic parameters to change before you can run Tigase for high volume and high load tests.
I don't know how to change it or whether it is needed on other systems (MS Windows?).
Thank you for the prompt
Thank you for the prompt reply,
I completely agree on your point about performance comparison.
At the same time we also wanted to have an understanding of the limitations of tigase server as to how many connections can be handled concurrently and other such things(specific to the hardware that we are using).Based on your suggestions about the Tsung test scripts for the load tests we have setup a Tigase server and are using tsung for testing it.
Our main aim to test was to identify
1. health monitoring perspective on the server side
resources during our test run.(Is there any tool
or add on available that will provide a better
understanding of the reources usage on server
side)
2. How many users can be simulated from a single
client machine and how do we reach that limit?
Currently We are facing problems as following(picked from the tigase.log file):
" ConnectionOpenThread.run() SEVERE: Server I/O error.java.io.IOException: Too many open files at sun.nio.ch.ServerSocketChannelImpl.accept0(Native Method)
at sun.nio.ch.ServerSocketChannelImpl.accept(ServerSocketChannelImpl.java:145)
at tigase.net.ConnectionOpenThread.run(ConnectionOpenThread.java)
at java.lang.Thread.run(Thread.java:619) "
It would be very helpful if you can help us out of this.
Thanks
SR
Performance comparison is
Performance comparison is always a very tricky task. It is very difficult to make sure tests are executed under the same conditions - hardware, database content, load, traffic shape.
I use Tsung to test the Tigase server. These tests were executed over the Tigase server using Tsung too. Please have a look at the tsung scripts and all the details.
I don't offer any advice for testing performance comparison. I am obviously biased towards Tigase software and all my suggestions could be unfair for other servers.
I am pretty sure there is no single ultimate winner in all categories. You might find one server particularly at some things while other can be better in others.
What is good about Tigase is that it is very flexible. If you find it not quite suited for your use case it is very easy to adjust it to ideally fit to your needs.
In any case, if you need any help with setting things up, testing or anything else you count on me.
Thanks for the reply, We
Thanks for the reply,
We wanted to identify which XMPP server to use & which would deliver the best performance. So in our expected load conditions we wanted to subject some servers to this load(using Tsung) and observe the performance of the servers.as Since we are still in test phase we are planning to evaluate tigase using Tsung against the load we expect & observe the performance throughput delivered.In such a case what would be the suggested approach for testing performance comparision?
Thanks in advance,
SR
There are many ways you can
There are many ways you can run such tests. You can use either one of available load testing tools: Tigase Test Suite or Tsung or you can write your own software for this.
I also heard of using dummy AJAX client for this. let's assume you have a busy website. You could put an invisible AJAX client on each page and users browsing your website would run the web client for you.
Particular approach depends on your specific case and your needs. There is no general solution appropriate for all cases.
Post new comment