Hi Michelle, Thanks again for the quick reply.
Regarding the hyperthreading: in fact 196% in user space of out 200% is different that 98% on the whole system. The "top" command shows two categories of measurements: on the whole (this is the output I pasted, with: 80.2% CPU in userspace, and 5.9% idle, while in the same time, top showed 196% out of 200 , for the tomcat process). I cannot use Axis2 but I can use Tomcat6.x. For the moment I am using Tomcat 5.5. I managed to eliminate IO operations (logging in tomcat) and now I am using another system monitor (mpstat from systat package); this one is a little bit better and tells me the IO percents the CPU has: it is now 0 %. However, the idle times are now up to 10% of the CPU, and now I have no idea why. The CPU waits because it was no data to process, but the receiver queue is empty on the network card, so the data is coming ... somewhere on the way. I tried to tune again the heap size (to very high and very low), the connection timeout in tomcat is now 0. What could I have missed ? Thank you for the help, Gabriela On 6/26/07, Michele Mazzucco <[EMAIL PROTECTED]> wrote:
Gabriela, please see my comments inline. On Tue, 2007-06-26 at 16:32 +0200, Gabriela Gheorghe wrote: > Already done the 1rst thing: increased the pool size from JMeter. The > results do not differ much, and this is my dilemma. You see, there are > limits when increasing the pool size of requests, depending on the > Tomcat configuration (no of active threads, request queue lenght , for > example), and it was the first possible think to do when I noticed > the server could do more. As I said in my previous post there are 2 thread pools: 1 - the tomcat connector thread pool 2 - the axis2 thread pool: your requests are processed using threads coming from here > > I changed and studied the heap size and thread pool size in Tomcat, > but nothing new happened. The problem is: I have only 196 out of 200% > CPU occupied and I cannot see why. The "top" command shows me > I don't get this point. How can you get 196%?, is it because of the hyper-threading (and so it's actually 98%)? If so, don't you think it's enough ;)? > Cpu(s): 80.2%us, 11.7%sy , 0.0%ni, 5.9%id, 0.0%wa, 0.5%hi, 1.8%si, > 0.0%s > > which means that some 5.9 of the processor is idle, while in Jmeter I > have 100 threads running in a loop making requests, and in Tomcat I > have 150 threads able to run simultaneously. Netstat -t shows me the > receiver queue is empty, so everything is being handled, but still, a > part of the processor is resting. > Which version of Tomcat are you running?, can you try Tomcat 6.x with NIO connector? Michele > Why ?! > > [Thank you for the responses] > > -- > Kind regards / Freundliche Gruesse, > Gabriela Gheorghe > > > Email: [EMAIL PROTECTED] / [EMAIL PROTECTED] > > On 6/26/07, Michele Mazzucco < [EMAIL PROTECTED]> wrote: > On Tue, 2007-06-26 at 10:22 +0200, Gabriela Gheorghe wrote: > > Thanks for the reply! > > > > The problem is not caused by I/O, because my web services do > no > > operations on the disk. The client is a JMeter instance > and the CPU > > load on the client machine is almost nothing (less than 8-10 > %). > > > Try to increase the number of threads in JMeter. > > > So what I am looking for is for tuning some other parameters > in > > Tomcat, beside heap size and connection timeout. I am still > loking for > > the explaination. > > > Actually they do if services return results to JMeter. You > could also > try to increase the tomcat/axis2 thread pool size. > > > Michele > > > All the best, > > Gabriela > > > > On 6/25/07, Michele Mazzucco < [EMAIL PROTECTED]> > wrote: > > > > On 25 Jun 2007, at 19:33, Gabriela Gheorghe wrote: > > > > > The problem is that I cannot run as many / > time-consuming > > requests > > > so that > > > the processor on the server reaches 100% CPU load. > It only > > reaches > > > 90%, so > > > the measurements for throughput that I need to > obtain by > > this testing, > > > cannot be too relevant; I want 100% of the CPU > working. > > > > > Are you sure that the bottleneck is not the client? > > > > > So my question would be - why is 10% of the CPU > idle ? > > > > Couldn't it be because of I/O? > > > > > > > > Michele > > > > > > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: > [EMAIL PROTECTED] > > For additional commands, e-mail: > [EMAIL PROTECTED] > > > > > > > > > > -- > > Kind regards / Freundliche Gruesse, > > Gabriela Gheorghe > > > > Email: [EMAIL PROTECTED] / [EMAIL PROTECTED] > > > --------------------------------------------------------------------- > 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]
-- Kind regards / Freundliche Gruesse, Gabriela Gheorghe University of the Federal Armed Forces Munich Fakultät für Informatik Werner-Heisenberg-Weg 39 85577 Neubiberg, Germany Email: [EMAIL PROTECTED] / [EMAIL PROTECTED]
