Hi! I have a question on what I might be doing wrong when commits of say 12 new files in a total of 34 deep directories (we're talking Java here) - totalling 44kiB - takes about 30 seconds to commit?
Sorry if this post is a bit long, I guess I got carried away.. I'm using Mac OS X 10.5.6 on a MacBookPro with 2 GB of RAM and get slow commits both with in Eclipse and on the command line. Eclipse is using Subversive using SVN/1.5.2 SVNKit/1.2.0, and the command line: : st...@mira ~;svn --version svn, version 1.5.2 (r32768) compiled Sep 22 2008, 12:24:00 The numbers here are from tests using the command line tool svn with - m "test" (or similar), and by uploading freshly the files and directories contained in http://soiland-reyes.com/stian/scrap/t2platform-testhelpers.zip. I used the unix tool "time" to record the real time spent. I've tested this from home today on an 5,550/1,178 ADSL connection in UK, but had almost the same numbers yesterday at work with a 100 MBit university connection. As for my limited upload bandwith affecting the results, doing a naive recursive scp to a university server takes 8.6 seconds, and scp-ing an uncompressed tar took 3.5 seconds. So still we have about 25 seconds of unexplained slowness. I tried this both for our project "taverna" ( https://taverna.googlecode.com/svn/taverna2/trunk/t2platform-testhelpers/ ), and as a test also in my private project "perfs" ( https://perfs.googlecode.com/svn/contrib/test/ ) in case Taverna's high revision (2000s) number affected it (perfs has around 100s), but I seem to get 29-31 seconds in both cases. I also tested upload to one of our own subversion servers in the university using svn+ssh - which took 7.2 seconds. To remove any doubts about the protocol I also tested against an https repository (https://s11.no/i/spine/t2platform-testhelpers/) on a server in Norway (powered by Apache 2 and an old subversion 1.4.0, with revisions currently at 659, backed by fsfs), this took 11 seconds. However, to this server the recursive scp test takes 22 seconds, and tar-scp takes 10 seconds. Is it latency issues affecting the svn commits to Google? On attempting to commit to the perfs project on googlecode with svn 1.4.0 logged in on this server in Norway, it took 50 seconds. In all of these cases it seems the client is spending most of it's time in the delay after all the dots have been printed in: Transmitting file data ........... and the result: Committed revision 185. For one of the test from my laptop against https://perfs.googlecode... , I recorded the times of these steps by observation. Wed 4 Feb 2009 17:26:58 WET Adding t2platform-testhelpers Adding t2platform-testhelpers/pom.xml .. Adding t2platform-testhelpers/t2platform-testhelpers-h3-impl/ src/main/resources/net/sf/taverna/platformtest/impl/ FooDataBean.hbm.xml Wed 4 Feb 2009 17:27:12 WET Transmitting file data ........... Wed 4 Feb 2009 17:27:18 WET Committed revision 187. Wed 4 Feb 2009 17:27:30 WET So in this case the "adding" took 14 seconds, "transmitting" 6 seconds, and 12 seconds were spent after "Transmitting file data" before "Committed" appeared, totalling of 32 seconds. Repeating this test from the university server took 27s, where: adding 14s, transmitting 5s, committing 9s. I believe "adding" is also doing the actual directory creation, because in one case svn commit failed: Adding t2platform-testhelpers svn: Commit failed (details follow): svn: Server sent unexpected return value (405 Method Not Allowed) in response to MKCOL request for '/svn/!svn/wrk/76e089ca-6659-41be-8f3b- b0bbec65e32b/contrib/test/t2platform-testhelpers' (I did a svn up (all the files got the status E), svn rm, svn commit to clean up - but still a bit worrying) As our main project I'm trying to commit is about 3000 files, which now takes "forever" (at least it did not finish in 60 minutes from work), I wonder what I can do differently to improve the speed. As a workaround I've been committing module by module, but as each takes from 30 seconds to 5-10 min this is a big pain, and I don't feel I can really recommend my team to use Subversion hosting by Google if these are the speeds we would have to live with.. If it's any comfort, performing the same test over https to Subversion at SourceForge takes 43 seconds, whereas 35s is "adding", 8s transferring, and only 2s committing - however doing it using CVS at SourceForge and Eclipse takes only 15s to add/commit - and that's our current solution that we now are not very sure if we should move away from.. -- Stian Soiland-Reyes --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "Hosting at Google Code" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/google-code-hosting?hl=en -~----------~----~----~----~------~----~------~--~---
