I got side tracked for a few days, but I finally got time to run some benchmarks with JProbe, as well as repeat a few of my original tests to verify the performance.
The tests were ran on the following systems. System I - 450mhz PIII, Win2K, 512mb RAM programs running - mcafee, aim Server - tomcat 4.0.3 additional connector - coyote beta 3 System II - 600mhz PIII, win2k, 256mb RAM programs running - none Server - tomcat 4.0.3 additional connector - coyote beta 3 load testing application Latest JMeter from jakarta Network - LAN JProbe 30 - added settings for Tomcat 4.0.3 test run - 2 threads 100 iterations The results below is the total number of Tag objects garbage collected for the entire duration of the test. Garbage collection was trigger right before and immediately after the run. The GC filter was set to "taglibs*". The system used was #2 with 600mhz cpu and no other apps running besides JProbe and Tomcat. JProbe GC results --------------------------------- include directive --------------------------------- IfTag - 11,200 WhenTag - 4,200 ExprTag - 2,800 SetTag - 1,400 OtherwiseTag - 1,200 ChooseTag - 1,200 Logger - 16,600 Total - 38,600 General memory/GC behavior - GC spiked much more frequently than pages using action include. The duration between GC was approximately half that of action include. action include -------------------------------- IfTag - 11,00 WhenTag - 2,800 ExprTag - 2,600 Logger - 16,600 Total - 33,000 General memory/GC behavior - GC frequency was less and the interval between GC was twice as long as pages using include directive. It generally took 2-4x more page hits to reach the heap max. scriptlet page -------------------------------- no tags were garbage collected, since none were used. for the duration of the test, GC was trigger twice. In comparison, include directive for the duration of the test was called on average 30 times. action include called GC approximately 15 times. The results below were average from 3 runs of each setting. The machine running JMeter was a separate system. Only the HTML was retrieved. No images or other files were included in the calculation of the response times. Test performed on system #2 (600mhz, 256ram) were not heap optimized, since it is pointless with 256mb of ram. All times are in milliseconds. include directive - 1 thread 1000 ---------------------------------- httpconnector ave - 32 cpu usage - 45-80% coyote connector ave - 107 cpu-usage - 30-100% include directive - 2 threads 500 ---------------------------------- httpconnector ave - 403 cpu usage - 50-100% coyote connector ave - 378 cpu usage - 40-100% include directive - 4 threads 250 ---------------------------------- httpconnector ave - 974 cpu usage - 70-100% coyote connector - 474 cpu usage - 50-100% include directive - 8 threads 125 ---------------------------------- httpconnector ave - 2802 cpu usage - 100% coyote connector ave - 1660 cpu usage - 60-100% include directive - 16 threads 65 ---------------------------------- httpconnector ave - could not finish cpu usage - 100% coyote connector ave - 2994 cpu usage - 80-100% action include - 1 thread 1000 ----------------------------------- httpconnector ave - 48 cpu usage - 40-70% coyote connector ave - 49 cpu usage - 40-60% action include - 2 threads 500 ----------------------------------- httpconnector ave - 84 cpu usage - 55-95% coyote connector ave - 86 cpu usage - 40-80% action include - 4 threads 250 ----------------------------------- httpconnector ave - 190 cpu usage - 70-90% coyote connector ave - 175 cpu usage - 50-90% action include - 8 threads 125 ----------------------------------- httpconnector ave - 508 cpu usage - 70-100% coyote connector ave - 372 cpu usage - 60-95% action include - 16 threads 65 ----------------------------------- httpconnector ave - 1056 cpu usage - 90-100% coyote connector ave - 808 cpu usage - 70-100% action include - 32 threads 63 ----------------------------------- httpconnector ave - 2646 cpu usage - 100% coyote connector ave - 1036 cpu usage - 80-100% scriptlet page - 1 thread 1000 ----------------------------------- httpconnector ave - 22 cpu usage - 0-20% coyote connector ave - 21 cpu usage - 0-16% scriptlet page - 2 threads 500 ----------------------------------- httpconnector ave - 30 cpu usage - 20-40% coyote connector ave - 28 cpu usage - 0-25% scriptlet page - 4 threads 250 ----------------------------------- httpconnector ave - 49 cpu usage - 20-40% coyote connector ave - 47 cpu usage - 20-35% scriptlet page - 8 threads 125 ----------------------------------- httpconnector ave - 97 cpu usage - 20-40% coyote connector ave - 97 cpu usage - 20-40% scriptlet page - 16 threads 65 ----------------------------------- httpconnector ave - 179 cpu usage - 20-45% coyote connector ave - 171 cpu usage - 20-40% scriptlet page - 32 threads 63 ----------------------------------- httpconnector ave - 250 cpu usage - 30-60% coyote connector ave - 257 cpu usage - 20-40% scriptlet page - 64 threads 35 ----------------------------------- httpconnector ave - 445 cpu usage - 30-80% coyote connector ave - 706 cpu usage - 40-90% Tests running on system #1 were all run with the same heap setting of "-Xms50m -Xmx280m". No additional optimizations were used in the test. include directive - 1 thread 1000 ----------------------------------- httpconnector ave - 24 cpu usage - 35-50% coyote connector ave - 23 cpu usage - 30-45% include directive - 2 threads 500 ----------------------------------- httpconnector ave - 35 cpu usage - 50-100% coyote connector ave - 40 cpu usage - 50-90% include directive - 4 threads 250 ----------------------------------- httpconnector ave - 241 cpu usage - 75-100% coyote connector ave - 119 cpu usage - 65-100% include directive - 8 threads 125 ----------------------------------- httpconnector ave - 962 cpu usage - 100% coyote connector ave - 403 cpu usage - 65-100% include directive - 16 threads 65 ----------------------------------- httpconnector ave - 3948 cpu usage - 100% coyote connector ave - 1173 cpu usage - 90-100% include directive - 32 threads 63 ----------------------------------- httpconnector ave - could not complete cpu usage - 100% coyote connector ave - 7907 cpu usage - 100% action include - 1 thread 1000 ----------------------------------- httpconnector ave - 34 cpu usage - 60-80% coyote connector ave - 35 cpu usage - 45-60% action include - 2 threads 500 ----------------------------------- httpconnector ave - 59 cpu usage - 70-90% coyote connector ave - 54 cpu usage - 55-80% action include - 4 threads 250 ----------------------------------- httpconnector ave - 118 cpu usage - 75-95% coyote connector ave - 112 cpu usage - 75-95% action include - 8 threads 125 ----------------------------------- httpconnector ave - 221 cpu usage - 80-95% coyote connector ave - 201 cpu usage - 75-100% action include - 16 threads 65 ----------------------------------- httpconnector ave - 457 cpu usage - 85-100% coyote connector ave - 423 cpu usage - 80-100% action include - 32 threads 63 ----------------------------------- httpconnector ave - 1099 cpu usage - 100% coyote connector ave - 860 cpu usage - 95-100% One obvious result of these benchmarks is memory is more crucial than CPU speed for site using lots of tag with JSTL. Using scriptlet pages, memory is less of an issue, but once GC kicks in, performance begins to degrade rapidly. Sites that wish to use lot of tags with include directive should consider having lots of memory. Not having enough memory seriously impacts the scalability of a website using JSP tags. >From the JProbe garbage collection data, using tag pooling is very desireable for deployments that use lots of tags. Re-using tag classes will dramatically decrease the amount of tag classes GC-ed and therefore improve resposne time and concurrency. peter __________________________________________________ Do You Yahoo!? Yahoo! Greetings - send holiday greetings for Easter, Passover http://greetings.yahoo.com/ -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
