have you tried monitoring the CPU and IO usage of the system during the test?
In the past, when I stress test an application, I monitor the cpu and io, to determine which part is getting maxed out first. For example, if I was serving up static pages, the first thing to mak out is the IO, so even though tomcat could handle greater loads, it can't because of the IO. What I usually do is identify the bottlenecks in the application by testing each part separately. Once I know the max through of each component, I ran the stress test with JMeter. If you haven't measured each piece individually, it will be difficult to figure why the application is slow. hope that helps peter On Jan 26, 2008 11:32 AM, Ali Ok <[EMAIL PROTECTED]> wrote: > Peter, thats ok, maybe some day we can get that much hit :) > > What if someone make so much requests and confuse the server? > Does Tomcat have an prevention for this situation? Or is it beyond the > scope? > > David, I have already read all of resources you sent. > > Invariably performance issues are rarely a result of the web container > > (Tomcat). You need to look at the developed software (dot).war that is > > deployed on TC (use JMeter or some recognized software testing tools). > > > > > * You are hitting your TC with 30000 transactions with a ramp up speed of > 10 > > seconds so you are at 3000/sec. If as you say the web container is not > > handling this you still need to look at what you web application is > doing. > > > > I dont talk about my application. It may contain bugs or incorrect logic, > blaming Tomcat is not my intention. Many people thank developers and I > admire Apache SF. > > I make that test of mine with "Shuffle Example" comes with Tomcat (XXX > /examples/jsp/jsp2/jspattribute/shuffle.jsp). And it is a simple JSP page, > cannot contain bugs :) > > * What is your planned network topology once you go to production? This > > question naturally leads to what is your hosting options? If you have > a > > datacenter and have configured and installed your own servers is the > best. > > Next option is to build your servers and then co-locate. The least > > advantageous option is renting servers (serverbeach.com etc.). If you > are > > hosting locally what is your upstream provider? (fat pipe) and type of > > connection: T1, T3, OC1, OC3 etc. > > * Type of scaling (horizontal vs vertical) > > > > My purpose is finding out the limits of Tomcat with constant resources at > the moment. Then I will continue with these. > > > I just want to have an idea in general. However its better to give > details. > > OS: Debian 4.0 r0 on VmWare > RAM assigned to virtual machine: 1 gb > CPU: Intel Core 2 Duo, 2 Ghz etc. > Tomcat: 6.0.13 > > > My starting script: > > /home/kullanici/Desktop/programlar/jdk1.6.0_03/bin/java -Xmx256m -Xms128m > - > Djava.util.logging.manager=org.apache.juli.ClassLoaderLogManager - > > Djava.util.logging.config.file=/home/kullanici/Desktop/programlar/Tomcat_6_win/conf/logging.properties- > > Djava.endorsed.dirs=/home/kullanici/Desktop/programlar/Tomcat_6_win/endorsed-classpath > > :/home/kullanici/Desktop/programlar/Tomcat_6_win/bin/bootstrap.jar:/home/kullanici/Desktop/programlar/Tomcat_6_win/bin/commons- > logging-api.jar - > Dcatalina.base=/home/kullanici/Desktop/programlar/Tomcat_6_win - > Dcatalina.home=/home/kullanici/Desktop/programlar/Tomcat_6_win - > Djava.io.tmpdir=/home/kullanici/Desktop/programlar/Tomcat_6_win/temp > org.apache.catalina.startup.Bootstrap start > > > Thank you, Peter and David > > > > 2008/1/26, David Brown <[EMAIL PROTECTED]>: > > > > Hello Ali, there are no absolute benchmarks for what you are looking > for. > > The central theme to any performance questions invariably lead to the > <A> > > word (Architecture). You need to evaluate you overall architecture from > a > > high level perspective. With this said the questions then are: > > * What is your planned network topology once you go to production? This > > question naturally leads to what is your hosting options? If you have > a > > datacenter and have configured and installed your own servers is the > best. > > Next option is to build your servers and then co-locate. The least > > advantageous option is renting servers (serverbeach.com etc.). If you > are > > hosting locally what is your upstream provider? (fat pipe) and type of > > connection: T1, T3, OC1, OC3 etc. > > * Type of scaling (horizontal vs vertical) > > * Invariably performance issues are rarely a result of the web container > > (Tomcat). You need to look at the developed software (dot).war that is > > deployed on TC (use JMeter or some recognized software testing tools). > > * I have worked on high volume financial web applications that are > running > > at 3 to 4000 transactions/sec. A transaction is end-to-end a round-trip > time > > starting with the HTTP connection, processing and connection to the > backend > > DB, query results returned and subsequently a results web page > displayed. > > This is a HTTP get, put or post transaction. > > * You are hitting your TC with 30000 transactions with a ramp up speed > of > > 10 seconds so you are at 3000/sec. If as you say the web container is > not > > handling this you still need to look at what you web application is > doing. > > > > Ultimately, using JMeter you need to look here: > > http://wiki.apache.org/jakarta-jmeter/ > > > > A expert in this area is Peter Lin: > > http://tomcat.apache.org/articles/performance.pdf > > > > The JMeter has specific JSF testing reading: > > http://wiki.apache.org/myfaces/PerformanceTestingWithJMeter > > > > Ali Ok wrote .. > > > Thanks David, > > > > > > I mean, if I make 30000 requests in a very short time (about 10 > > seconds); > > > Tomcat does not respond. > > > I read books, tutorials, faqs and threads at maling list about Tomcat > > > tuning. But I couldnt find an example server.xml file used in > production > > or > > > real test results. > > > > > > So I cant understand if 30000 requests in 10 seconds is normal or not. > > > > > > > > > > > > 2008/1/26, David Brown <[EMAIL PROTECTED]>: > > > > > > > > Hello Ali, please find included below a link URL that addresses the > > JSF > > > > performance issue. A much more rigorous test would be to use the > > JMeter > > > > distributed testing using the JMeter server. HTH, David. > > > > > > > > Ali Ok wrote .. > > > > > Hi, > > > > > > > > > > We are building a web application with JSF. Last day I tested it > > with > > > > > JMeter. Results are bad (I guess). > > > > > > > > > > Then I tried to send 30000 requests with JMeter to "Shuffle > Example" > > in > > > > > Tomcat's examples directory with a limited size of (256 MB I > think) > > > > memory > > > > > resource given to Tomcat. This "Shuffle Example" does not query > > database > > > > or > > > > > does not make complicated operations as you know; it is very > simple. > > > > > > > > > > Question is, what should I expect? Does it have to respond all > > requests? > > > > Or > > > > > is it normal to throw an exception about "Too many open files" (I > > use > > > > NIO > > > > > connector) and finally OutOfMemoryError and parachute-thing? > > > > > > > > > > After I solve this, I can go on to JSF application testing. > > > > > > > > > > > > > > > I couldnt find documents enough about this issue. Can you send me > > some > > > > > links? > > > > > > > > > > Thanks in advance. > > > > > > > > > --------------------------------------------------------------------- > > > > To start a new topic, e-mail: users@tomcat.apache.org > > > > To unsubscribe, e-mail: [EMAIL PROTECTED] > > > > For additional commands, e-mail: [EMAIL PROTECTED] > > > > > > > > > > > > --------------------------------------------------------------------- > > To start a new topic, e-mail: users@tomcat.apache.org > > To unsubscribe, e-mail: [EMAIL PROTECTED] > > For additional commands, e-mail: [EMAIL PROTECTED] > > > > >
