Hello Victor, We're about to publish beta2 that does address these issues.
We did turned off synchronous writes to the database like native Subversion does, plus other improvements were made and there is still space for it. For instance, in memory transaction journal might be used, or we could combine transactions or run them in parallel as you've suggested. Currently checkout performance should be comparable to that of the native client and we will continue with performance improvements in the upcoming versions of SVNKit. Thanks! Alexander Kitaev, TMate Software, http://subgit.com/ - Svn to Git Migration! http://svnkit.com/ - Java [Sub]Versioning Library! http://hg4j.com/ - Java Mercurial Library! http://sqljet.com/ - Java SQLite Library! On 26 February 2012 21:06, Victor Borisoff <[email protected]> wrote: > Hi all, > > Firstofall I want to express my gratitude as I see the svnkit 1.7 been > developed at such a tight schedule. It is a great product. > > I have tried to use svnkit 1.7 beta 1 on our codebase repository (1.8Gb of > source files sitting on the SVN server 1.7). I noticed issues with > performance of the svnkit 1.7 executing on Windows workstation. It does > complete checkout in 28 minutes. This is much slower than the native linux > svn client (full checkout in 5 minutes). > > I have noticed that the svnkit checkout spends most of the time in executing > JetSQL transactions. Is there a ways to speed up this access? Maybe I can > turn off synchronous write? > > Checkout of each file consists of four JetSQL transactions. Is it possible > to combine all or some of these transactions? > > I tried a simple way to speed it up by splitting the socket reading and > JetSQL command execution into two concurrent threads (attached is the > modified SVNEditModeReader.java). Would it be possible to put this change > into the official version of the SVN kit? > > > > Thanks a lot, > > Victor Borisoff
