Hey Yes our struts uses the tiles plugin also. And the page load times are as a result of running on the server itself unfortunately. I am using the JTDS driver which apparently is one of the fastest SQL Server 2000 implementations for db connectivity.
One of the things about our system is that every click the user makes is a search to the CMS for content that belongs to that page. This could be considered sub-optimal. For instance, I could cache results per 15 minutes or something, but content changes often so I thought I would leave it dynamic. I spent some time running JProfiler against the app. The accumulated CPU and memory seemed to ultimately come down to these search calls to the CMS especially under load. Per request I have to instantiate the vendors' bean which is an interface to the CMS server socket and is responsible for giving back all the data from the calls. I then simply package up what I need in data transfer objects and ship them to the JSPs. The JSPs are highly configurable in terms of skinning and so on and have calls to the file system to check if images are present and so on. I think I could cache that too. In fact looking at timings in the logs shows the page start to end is sub 1 second, the rest (user click -> page process start) I think is caught up at the search but even so that is generally about 0.8s. Pages are taking about 0.1 and my cached implementation of nav takes 0.06 or so to load. But overall the pages under load feel like they take about 2 seconds and also I get the feeling tomcat is recompiling something it really doesn't need to - that was why I was interested in the development flag in server.conf and whether I should turn that to false. A fair amount of CPU and memory was going to Log4J also. I removed a lot of our unnecessary debugging statements and this too helped. I guess it comes down to the CMS load but any tips from others on the list are so very welcome, Cheers, ADC -----Original Message----- From: LILES, DAVID (CONTRACTOR) [mailto:[EMAIL PROTECTED] Sent: 08 April 2004 17:22 To: Allistair Crossley Subject: RE: Tomcat configuration tuning I have seen a couple postings regarding poor performance with Tomcat recently and thought I'd respond because we are using IIS5 with Tomcat 5 and SQL 2000 as well. We have a Struts site but our difference is that we are using Tiles for our presentation... haven't really looked into the performance differences between Tiles and non-Tiles display rates.... As for our database connectivity... we are using an internally developed connection pooling class (no JNDI), but have recently started migrating to Hibernate. It is interesting hearing how long it is taking for the pages to be displayed considering we are on a similar set-up and have not noticed this type of problem. We have not made any special optimization setting changes, we basically did an out of the box set-up.... Is this occurring when running the app locally or on the server? I would be curious to know what is the resource load on the box running the app, how is the connectivity (bandwidth) to the box, has any sort of optimization tools been run against the code to check for bottle-necks? Not sure that I can provide much value technically, but feel free to ask.... -Dave -----Original Message----- From: Allistair Crossley [mailto:[EMAIL PROTECTED] Sent: Thursday, April 08, 2004 11:00 AM To: Tomcat Users List Cc: [EMAIL PROTECTED] Subject: RE: Tomcat configuration tuning Hi Matt, Welcome to the list from me. It's funny because you have just listed precisely the same server setup as we are launching our new tomcat based app to this weekend and I have been having nightmares about performance. We have a system like this; user -> IIS -> JK2 ISAPI -> Tomcat 5.0.19 -> --> Content Management System Bean --> SQL Server 2000 --> SQL Server 2000 via JNDI resource setup Our application is built on Struts. We found performance to be very poor under a load of 20 users of 3 iterators of 20 page requests (13 secs per page) avg pages per secon 1.1!! Looking into it revealed too much XML, too much database connectivity and too much dynamic rending on the pages. I came up with a caching strategy for our navigation as well as removing all XML to database calls instead (you may have seen my last post) Along with other things I have managed 10 pages per seconds and 1.5 seconds per page under the same load test. I'm still not entirely happy and wonder if I am missing any confoguration type stuff to make my app run faster either in tomcat, jk2, iis and so fourth. I run tomcat as a service with -Xms128 -Xmx384. I would be very interested if you have time to hear of any useful quick tips you found can optimise via configuration - maybe even seeing your server.conf and anything else, perhaps any ideas about code optimisation and also how complex your system is (db accesses or other system integration). What are your page response times under load and so on. It's a big ask but you have a very similar hardware setup and I would like to maximise my use of it. Cheers! ADC -----Original Message----- From: Matt Woodings [mailto:[EMAIL PROTECTED] Sent: 08 April 2004 16:50 To: [EMAIL PROTECTED] Subject: Tomcat configuration tuning Hi guys, I am at stage where I have all my various components talking to one another (Apache http server 2.0.49 -> Mod JK2 -> Tomcat 4.1.30) and running nicely with each other, but the loads on the server are increasing rapidly in our production environment ( ~2000 logins/day and peak concurrency of ~75 currently ) and I want to get the best possible configuration for my server. (Dual 1GHz, 1GB Ram, Win 2k, SQL Server + Web/Apps servers). Now, ultimately this is up to me, but I am unsure of a few relationships and variables used in the configuration files. The settings I have work, but I want them to be optimal and more importantly I want to understand them, so any help on the following would be greatly appreciated. 1) server.xml/Connector/bufferSize. Is this allocated for every users session until it is terminated, or is it a peak size that could be allocated? 2) workers2.properties/shm.slots. Does this equate to the server.xml/Connector/maxProcessors? or are these fundamentally different? I have attached my fully working configuration below for those who are interested, as I noticed a lot of people in the archives looking for working examples using JK2. Many thanks, Matt P.S. I am new to the list but I have been using Tomcat for many years and love it, and if I can help with anyones queries, from the trivial to complex, I would be only too happy to do so. workers2.properties [logger] ############################################################## #### LOGGER #### ############################################################## # Log level. Supported: EMERG, ERROR, INFO, DEBUG level=ERROR # Log file. XXX you may be able to change this at runtime, to # implement rolling. file=${serverRoot}/logs/jk2_error.log [config] ############################################################## #### CONFIG #### ############################################################## # Location of the workers2.properties file file=${serverRoot}/conf/workers2.properties # Set the debug level of the config component debug=0 # Set the debug level of the hidden env component debugEnv=0 [shm] ############################################################## #### SHARED MEMORY SETUP #### ############################################################## # Name of the file that will be mmapped to use as shared memory, # If set to 'anonymous' use the anonymous shered memory file=${serverRoot}/logs/shm.file # Number of shared memory slots. Set to the number of child # processes slots=378 # Use process memory instead of shared memory. Useful for single # child mpm's useMemory=0 [lb:lb_group] ############################################################## #### LOAD BALANCER (EVEN IF USED ON ONE MACHINE) #### ############################################################## # If set, jk2 won't touch the headers in case of error and will # let for example Apache present the # ErrorDocument via mod_alias. noErrorHeader=1 # noWorkerMsg=No workers available # noWorkerCode=503 # hwBalanceErr= # If all the workers are in the error state, probably by Tomcat # refusing any new connections due to the overload, you can set # the timeout forcing lb to wait that some worker becomes # available, instead of immediately returning error to the # client. This is very useful in situations with high peek # load. The timeout should be set to the maximum application # call time, but not less then 1 second. timeout=5000 # Number of attempts that lb will try on each worker before # giving up. attempts=6 # Time to wait before retrying to see if the worker came out of # the error state. Default = 60secs recovery=30 # Sessions stick to the same worker, 1=true 0=false stickySession=1 [channel.socket:localhost:8009] ############################################################## #### CHANNEL SETUP, LINKED TO LOAD BALANCER #### ############################################################## # Port where Tomcat is listening. It is automatically extracted # from the name - you shouldn't have to specify it explicitely. port=8009 # Remote host. You should use the name, no need to override it host=127.0.0.1 # If 1, only requests for existing sessions will be forwarded graceful=0 # ? keepAlive=0 # Socket timeout for sending and receiving (0=infinite) timeout=0 # Load balancing factor to use. At this moment, it'll be set on # the worker, but in future it should be possible to use lb on # a channel level. lb_factor=100 # loadbalanced groups to which this channel and the associated # worker will be added, multivalued. You need to set it only if # you have an advanced setup with multiple clusters. group=lb_group # Must match the JVM route on tomcat Engine, for load balancing tomcatId=hsInstance [ajp13:localhost:8009] ############################################################## #### AJP1.3 WORKER CONFIG #### ############################################################## tomcatId=hsInstance group=lb_group channel=channel.socket:localhost:8009 [uri:/hs/*] ############################################################## #### URI MAPPING, WHAT TO SEND TO TOMCAT #### ############################################################## group=lb_group [status:jk_status] ############################################################## #### JK STATUS GROUP #### ############################################################## info=Status worker, displays runtime information [uri:/jkstatus/*] ############################################################## #### JK STATUS CONFIG #### ############################################################## info=The Tomcat /jkstatus handler group=status:jk_status <FONT SIZE=1 FACE="VERDANA,ARIAL" COLOR=BLUE> ------------------------------------------------------- QAS Ltd. Developers of QuickAddress Software <a href="http://www.qas.com";>www.qas.com</a> Registered in England: No 2582055 Registered in Australia: No 082 851 474 ------------------------------------------------------- </FONT> --------------------------------------------------------------------- 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]
