Oleg and Roland, Thanks for the response. I am not aware of any special protections of the machine at the network level (is there a way to find out?), nor was I using SSL. I wonder why it makes use of SHA in this scenario.
Thanks, Ben -----Original Message----- From: Roland Weber [mailto:[EMAIL PROTECTED] Sent: Tuesday, 17 February 2004 9:50 PM To: Commons HttpClient Project Subject: RE: GetMethod Performance Hello Ben, Sorry, I got mixed up because yesterday there was another poster reporting performance problems when using SSL. So forget my comment about the secure socket factory. For some reasons, the slow machine seems to compute an SHA1 hash sum. If it's doing that for the 2 MB file, that would be an explanation for the difference in speed. Are there any special protections of the machine on the network level? If not, it might be worthwile to run your test on a much smaller file and to get a wire log for both machines to compare the actual HTTP messages that are transfered. cheers, Roland "Ben Wong" <[EMAIL PROTECTED]> 17.02.2004 11:10 Please respond to "Commons HttpClient Project" To: "'Commons HttpClient Project'" <[EMAIL PROTECTED]> cc: Subject: RE: GetMethod Performance I am running the same code on both boxes, even the same JAR file. I am not sure why crypto is involved as I was only retrieving content over HTTP. I used the HttpClient to download a zip file off a Web server in the LAN. It should take no time to download it. With the wire log, I have actually removed the log4j.properties file completely from the class path, so it shouldn't be affecting the performance should it? Thanks, Ben -----Original Message----- From: Kalnichevski, Oleg [mailto:[EMAIL PROTECTED] Sent: Tuesday, 17 February 2004 8:12 PM To: Commons HttpClient Project Subject: RE: GetMethod Performance Ben, The profiles obviously differ quite a bit. For instance the methods below that consume a significant portion of the execution time on Solaris platform take virtually no time on WinXP. Are they executed at all? Are you sure you execute exactly the same code on both platforms? Anyhow, as Roland said, the problem is clearly related to crypto algorithms on the Solaris platform. = Solaris ================================================ Interpreted + native Method 5.6% 18 + 6 java.math.BigInteger.mulAdd 3.3% 14 + 0 sun.security.provider.SHA.computeBlock 3.0% 13 + 0 java.math.BigInteger.addOne 2.8% 11 + 1 sun.nio.cs.StreamDecoder.read =========================================================== One additional note: please make sure that you have debug/wire log turned off when conducting performance measurement, otherwise you simply end up measuring the speed of console output Oleg --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
