http://www.lynn.edu/

Notes about the drawing

• Everything inside the blue box is virtualized on VMware, except for the storage. (I could be wrong about the storage, but I am trying to get our I.T. guy in here or I will followup and clarify.)
• All servers are Windows 2003 Servers.
• I don't think the storage is mounted via NFS.
• This drawing is slightly updated from an original provided by Cignix, with whom we consulted prior to our site launch in October 2008.
• Only Zope Instance 2 and Zope Instance 3 are part of the load balancing.

Notes about this setup

• The reason for the off-site proxy is emergency fail-over. We are in South Florida, threatened by hurricanes, among other things. If campus loses Internet, electricity, or we get slammed with traffic, we want to be able to have a low-bandwidth site at a distant server provide important information without waiting for DNS propagation.
• We went with a virtualized setup setup primarily on the recommendation of our I.T. department which is migrating many services to VMware. We were under a time crunch, and this was much faster than procuring hardware.

Background on the project

We are a small team of 3 web designer/developers plus 1 primary support staff member in the university I.T. department. None of us had prior knowledge of Plone, Zope or Python. The first time I heard of Plone/Zope was March 2008. In June 08 we were developing custom content types with the help of Cignix and we launched in October 08. Of course, I think we are still at the base of the Plone learning curve.

Performance

In the 13 months since we launched the site, we have had some performance issues, but I'm inclined to believe it is due to our lack of Zope and Plone knowledge, rather than virtualization.

• We have the two Zopes in the load-balancer auto-restarting twice a week due to Python RAM usage.
• We only recently enabled CacheFu and seem to have it dialed in ok, but I'm sure there is room for improvement.
• We also recently increased cache-size on Zope clients to 10,000.

Despite a few hiccups, this has been relatively stable setup for us, though we often wonder if we would notice a performance gain in a non-virtualized environment.

Occasionally we have an issue where one, then the other, of the load-balanced Zope clients will stop responding to web requests, but when we look at the RAM/CPU usage and log on to the server, everything seems fine. Usually the CPU is elevated, but not pegged. None-the-less, we have to kill the process and start it up again. These issues seem to come in bursts. We go weeks with no problems, then have a few incidents over the course of a week or two.

Virtual Server Specs for Plone

LynnWebZeo:

Server - VMware

CPU - 4 x 2.3GHz Intel Xeon

RAM - 1536MB

Storage - RAID5/Fiber Drives/iSCSI

LynnWebZope1:

Server - VMware

CPU - 4 x 2.3GHz Intel Xeon

RAM - 2048MB

Storage - RAID5/SATA Drives/CIFS Share

LynnWebZope2:

Server - VMware

CPU - 4 x 2.3GHz Intel Xeon

RAM - 2048MB

Storage - RAID5/SATA Drives/CIFS Share

LynnWebZope3:

Server - VMware

CPU - 4 x 2.3GHz Intel Xeon

RAM - 2048MB

Storage - RAID5/SATA Drives/CIFS Share

Physical Server Specs

Hosts are HP ProLiant G5, each with Quad Core Intel Xeon 2.33GHz and 20GB RAM

VMware Specs

VMware ESX 3.5 with VMware Virtual Center 2.5

Plone Specs

Currently: Plone 3.3.1, Zope 2.10.9

LynnWebZope2&3:

cache-size: 20,000

zserver-threads: 4