-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Jason,
On 3/25/2010 3:01 AM, Jason Brittain wrote: > Chris: there are no units on your results numbers, and I'm not seeing > any procedure you used, nor any configurations you used, so I'm not sure how > to interpret the numbers. I'd be happy to give you a quick explanation, while a complete writeup is still... on the back burner. Those Tomcat people keep putting out new releases and it takes a long time to run all the tests. I have yet to run keepalive versus non-keepalive (well, just the keepalive test) against all the connectors (plus httpd) AND Andre' asked about SSL, so I suppose I'll have to try that, too. Here's the deal: All numbers in the cells are effective transfer rate (in KiB/sec) over an 8-minute testing window: basically, I made as many requests as I could for 8 minutes straight to a single static file (file size listed in the left-hand column) and let ApacheBench tell me what the transfer rate was (which IIRC does not include HTTP headers, etc.: just the file content). It looks like Mark cherry-picked the results with this profile: keepalive=off, concurrency=40, Client VM I also did concurrencies (parallel client threads) of 1, 80, 160, and 200 (I think... I hadn't yet merged that data into my spreadsheet). It's all very repeatable using a set of scripts I wrote for this purpose. > It would be great to get more information about > how the benchmark was conducted, which HTTP client was used, and what server > hardware was used. - From my forthcoming (!) write-up: " These tests were performed on a modest machine with a single-core 32-bit microprocessor (see Appendix A for a complete description of the test hardware) and 1GiB RAM. Tomcat 6.0.20, tcnative 1.1.18, and apr 1.3.8 was tested on Sun's Java Virtual Machine 1.6.0_15_b03 (client and server JVMs were tested separately: see the individual tests for details). Apache httpd 2.2.12 was used for comparison. Both httpd and Tomcat were used in their default configurations where applicable (that is, no performance-oriented tuning was performed on either configuration). ApacheBench 2.3 was used to test transfer rates from each server configuration. The tests were run from the local machine to avoid network interference. " Unless otherwise specified, all software was kept in it's default configuration. That is, no tuning was performed on any of the components for these tests. > I did try to think up and benchmark > the most likely use cases for serving typical webapp content, but anyone can > say their webapp isn't like that. I stuck to static files because nobody cares what the performance of running a JSP relative to httpd is... since HTTP doesn't serve them :) > I'm happy to see that Chris' independent benchmark numbers help to show that > it is indeed a myth that Tomcat needs HTTPD in front of it in order to get > good performance serving static files. And, it's great to see benchmark > results for file sizes that I wasn't able to benchmark. I also intend to show what the overhead is of adding httpd "needlessly" in front of Tomcat. I suspect that it won't be that bad :) - -chris -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.10 (MingW32) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iEYEARECAAYFAkurqD0ACgkQ9CaO5/Lv0PBTTQCeNdqh/MEeFA0pdrlXtnWNC9qI ZY4AoLNyKI2RyhL64tcEoqDjzlVitqqY =iBpD -----END PGP SIGNATURE----- --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
