Re: https / ssl performance, 1.8 (answer: SVNPathAuthz Off)

Thanks so much for the replies; I had already set SVNAllowBulkUpdates to Prefer. But reviewing the description gave me another opportunity to RTFM ;)

Turns out the issue was path authorization; although I had read and understood we didn't want to use it, I missed the fact the default is "On" (What's up with that?). Turning it off solved the problem.

SVNPathAuthz On|Off|short_circuit

Controls path-based authorization by enabling subrequests (On), disabling subrequests (Off; see the section called "Disabling path-based checks"<http://svnbook.red-bean.com/en/1.7/svn.serverconfig.httpd.html#svn.serverconfig.httpd.authz.pathauthzoff>), or querying mod_authz_svn directly (short_circuit). The default value of this directive is On.

On 10/9/15 11:05 AM, Philip Martin wrote:

"Weatherby,Gerard" <gweatherby_at_uchc.edu><mailto:gweatherby_at_uchc.edu> writes:

It "works" but the performance makes it nearly unusable. Monitoring
the server using the top linux utility indicates many, many
invocations of "pwauth" for a single client request. The next thing I
was going to try was to install an sssd daemon on the server, but I
wanted to ask if there was anything else I was missing.

By "single client request" I assume you mean a high level operation such
as checkout, rather than an HTTP request such as GET. 1.8 is probably
using skel mode changing from a single large HTTP request to multiple
small HTTP requests. I think pwauth runs a separate process to
authenticate each HTTP request so it is not as efficient as other auth
methods. If you must use pwauth you might try disabling skel mode to
get behaviour more like 1.6:

http://subversion.apache.org/docs/release-notes/1.8.html#serf-skelta-default

--
Gerard Weatherby | Application Architect
Virtual Cell | Center for Cell Analysis & Modeling | UConn Health
NMRbox | Department of Molecular Biology and Biophysics | UConn Health
263 Farmington Avenue, Farmington, CT 06030-6406 uchc.ed