Hi Also to add usually thread related problems on linux fall into two categories (that i know off) a. Since every socket counts as an open file descriptor , the total number of allowed open file descriptors should be set to a high enough value. b. The stack space per process should be increased (using ulimit or something like that)
Again I dont know enough of unix to tell you what the commands actually but these are the problems you usually run into when configuring threading for a java app so you might read up on the two things above. regards deepak On Fri, Nov 26, 2010 at 5:52 AM, Deepak Shetty <[email protected]> wrote: > Thats funny - the parameters you see should impact garbage collection (on > Sun JVM) . > What JVM are you running (inclusive of version) - XX parameters may not be > supported on all versions / OS/ and are subject to change (and valid only > for SUN) - check the documents > > However they should have no impact on the number of threads you can spawn > (unless you are getting an Out of memory) - though yes they can have an > impact on performance. > > When you say cant start what errors do you get? Its possible you are seeing > some other issue on Linux > > regards > deeoak > > > On Fri, Nov 26, 2010 at 5:40 AM, Adrian Speteanu <[email protected]>wrote: > >> Hello, >> >> Need some advice on configuring jmeter shell script to run better on linux >> (debian/ubuntu in my case). >> >> Uncommenting these parameters in jmeter script has adverse affects: >> # SURVIVOR="-XX:SurvivorRatio=8 -XX:TargetSurvivorRatio=20%" >> # EVACUATION="-XX:MaxLiveObjectEvacuationRatio=20%" >> The JVM doesn't want to start with them on debian, although I ran >> jmeter.bat >> with them on default values or improved targets, on Windows XP, Vista, 7 >> stations: >> >> I didn't find an answer in previous threads, though this was discussed >> some >> time / some years ago. >> >> Still jmeter works without, BUT there is a noticeable performance penalty >> without them: with the changes I make in jmeter.bat, I can start more then >> 4000 threads on a Win XP desktop system, with a simple script; on the same >> hardware with ubuntu installed or on other more powerful machines with >> debian I can't start more then 2000 threads, same script of course (using >> GUI or not). Considering that XP has a max threads limit of 10.000 and >> that >> the server had 16g of ram, this upsets me, it should work better on linux. >> The script doesn't depend on an external application, just uses a lot of >> CPU >> and memory and I used to tune how jmeter runs, however I have encountered >> some problems with JVM in tests that ran for longer periods of time. >> >> Any suggestions? >> >> ---- >> >> just in case, here is what is in jmeter shell script: >> >> HEAP="-Xms512m -Xmx1512m" >> NEW="-XX:NewSize=64m -XX:MaxNewSize=128m" >> #SURVIVOR="-XX:SurvivorRatio=8 -XX:TargetSurvivorRatio=20%" >> TENURING="-XX:MaxTenuringThreshold=2" >> #EVACUATION="-XX:MaxLiveObjectEvacuationRatio=20%" >> RMIGC="-Dsun.rmi.dgc.client.gcInterval=600000 >> -Dsun.rmi.dgc.server.gcInterval=600000" >> PERM="-XX:PermSize=64m -XX:MaxPermSize=64m" >> DEBUG="-verbose:gc -XX:+PrintTenuringDistribution" >> DUMP="-XX:+HeapDumpOnOutOfMemoryError" >> (the rest is unchanged). >> >> Thanks & regards, >> Adrian S >> > >
