hi everybody, sorry to reopen such an old issue, but i have remembered some discussion threads about closed sockets and other socket issues (mostly when testing from a Windows station) and thought to bring something up: http://half-open.com/
on some versions of Windows, the tcpip.sys is not configured for load testing (the configuration is good for home users, as it limits flood possibilities and a number of other things, from what I understand). Increasing the number of possible "half-open" connections helps with connection closed or connection already in use issues, at least from what I have seen. i haven't encountered the "java.net.BindException" exceptions with the new jvm releases installed (with or without changing the number of half-opened connections). i've used the tool after trying these configurations, recommended by deepak: http://publib.boulder.ibm.com/infocenter/cmgmt/v8r3m0/index.jsp?topic=/com.ibm.eclient.doc/trs40019.htm (it's been helpful and necessary, also, but didn't eliminate all of the issues). if you have some more inputs on the matter... On 7/30/09, Deepak Shetty <[email protected]> wrote: > hi > netstat works on windows as well. Im not sure whether you'd get better data > with perfmon (same dealine pressures prevent me from trying to investigate > this properly). > > If you arent getting problems with 200 * 200 , and you need to simulate > heavier loads then can you distribute your jmeter test across multiple > machines? > > Someone just posted a script to load the CSV files into a table and run > reports, you could tweak it to ignore error samples > regards > deepak > > On Thu, Jul 30, 2009 at 11:00 AM, shaoxianyang <[email protected]> wrote: > >> >> Hi, >> >> Both my server and client(jmeter) machines are windows xp machines. >> >> Deepak, how do I monitor socket in windows? If you know? Otherwise, I >> guess I have to install cygwin to do the netstat (i know windows has it) >> and >> grep thing. >> >> As for max open file limit, I don't think mine has that problem, since >> windows does not have that limit, right? >> >> Since I need to collect and analyze some data soon, I won't be able to >> keep >> tracking this. I will only log the success result at this time. But >> since >> once there is failure, the result will be deviating too much. Shall I >> just >> trust the "median" out of all the result? Obviously, that average is not >> helping at all. Or there is other way to get more accurate result on how >> long each call takes? >> >> I will continue follow your instruction on tracking on this, once I am >> free >> on the current urgent task. I will keep you guys posted. >> >> Thanks. >> >> >> >> Shaoxian >> -- >> View this message in context: >> http://www.nabble.com/java.net.BindException-returned-from-http-sampler-tp24716020p24743685.html >> Sent from the JMeter - User mailing list archive at Nabble.com. >> >> >> --------------------------------------------------------------------- >> To unsubscribe, e-mail: [email protected] >> For additional commands, e-mail: [email protected] >> >> > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
