Setup
Pound/Router
Sun V60x running Debian/stable (2.4.something), iptables and
Pound-current. One external interface with a real, resolvable IP
address, one internal private interface onto a gigabit switch with the
backend servers.
pound.cfg:
Listen REAL.I.P.ADDRESS:80
Server 10
UrlGroup "/Images/.*"
HeadRequire Host ".*SITE.ncl.ac.uk.*"
BackEnd SITE-IMAGES.ncl.ac.uk,80,1
EndGroup
# Main cluster
UrlGroup ".*"
HeadRequire Host ".*SITE.ncl.ac.uk.*"
Session IP 1800
BackEnd backend1.ncl.ac.uk,22221,1
BackEnd backend2.ncl.ac.uk,22222,1
BackEnd backend3.ncl.ac.uk,22223,1
EndGroup
# Main cluster
UrlGroup ".*"
HeadRequire Host ".*SITE2.ncl.ac.uk.*"
Session IP 1800
BackEnd backend1.ncl.ac.uk,33331,1
BackEnd backend2.ncl.ac.uk,33332,1
BackEnd backend3.ncl.ac.uk,33333,1
EndGroup
--------------
Backend servers
3 x Sun V20z dual Opteron boxes, with Suse linux 9.1 (2.6.5) 64bit, each
running two separate ZEO 2.7.2 clients (backend1,2,3), python2.3.3
64bit, with the ZEO server on a E3500.
--------------
The sites being hosted are blazingly fast, and we are confident we can
support the amount of users expected for both... however, when the
developers are navigating the ZMI interface, there is a consistent,
measureable 5-second delay when browsing to any folder within the ZMI
view.
Viewing logs from pound and the 3 ZEO clients, the delay is a 5 second
gap from a manage_workspace request (with a 302 response), to the rest
of the page loading in. It is the same on all backend servers (we've
tried them seperately) and even on a new, blank Zope install inside the
cluster with an empty data.fs.
In desperation to the angry developers now ganging up on me, I have just
gone back and tried Pound-1.7 with the same compile switches
(--without-log, -O2) and config and it completely resolves the problem.
I'm baffled!
Did something break in the latest Pound-current release?
-John
-==========================================-
School of Medical Education Development
Faculty of Medical Sciences Computing
University of Newcastle
|
