On Thu, Dec 27, 2012 at 7:28 AM, RW <[email protected]> wrote: > On Thu, 27 Dec 2012 02:33:00 -0800 > Garrett Cooper wrote: > >> On Thu, Dec 27, 2012 at 2:30 AM, Chris Rees <[email protected]> wrote: >> > > >> > I recall a cluster administrator advising use of svn protocol >> > rather than http. Something to do with overheads. >> >> Yes, you're right. http opens a connection per-file, which is >> horrendous with server load and is very slow as HTTP isn't optimized >> for this purpose > > Presumably that's a limitation of the library or subversion because it's > not true of HTTP per se. Portsnap, for example, uses persistent > connections with pipelining.
An example of what I mean [1]: ======================= A checkout/update over svn:// is approx 4-12 times faster than over http(s):// . The factor depends on number of files/folders and file size . Apache is much slower on many small files because each file is a full http-request-response cycle. In Tortoise you can easily see the speed dropping: Checkout a large java projects If you transfer a huge .jar, transfer is going up, on getting source java files and creating package structure, it will drop. Also important is that svn checkout is slower on client than svn export and also eclipse(java) is much slower than tortoise/CMD. share|improve this answer ======================= Qualitatively we (my group) have seen a huge performance increase in our svnsyncs from freebsd.org using svn:// instead of http:// . We haven't measured how much the difference is quantatively though because a qualitative answer was sufficient for helping reinforce a business case for unfiltering that port with corporate. HTH, -Garrett 1. http://stackoverflow.com/questions/372569/subversion-protocol-performance _______________________________________________ [email protected] mailing list http://lists.freebsd.org/mailman/listinfo/freebsd-hackers To unsubscribe, send any mail to "[email protected]"
