I’m using MOSS in a WCM configuration with an authoring farm that deploys content to a Read-Only Internet farm. The authoring and Internet farms all consist of the same server hardware and use remote SQL 2005 clustered database instances. I’m using Dell 2950 servers with dual quad-core 2.6Ghz processors, 8GB Ram, SAN attached storage, 64-bit OS and 64-bit SharePoint.
The Authoring farm consists of the following:
Authoring Server 1: WFE for the 200 authoring users.
Authoring Sever 2: Central Admin, Application services (Excel, email, index, document, etc..)
The Internet farms consists of the following:
Internet Server 1: WFE – Public URL
Internet Server 2: WFE – Public URL
Internet Server 3: WFE – Internal URL of public URL, Application services (excel, email, docs, index, etc..) plus Central admin.
The Internet Servers 1 and 2 are fronted by dual F5 Big-IP 6400’s with Web accelleration and the SharePoint application pool template.
Why are we using this hardware configuration?
Hardware performance testing that led me to this configuration included running VMWare 32-bit configurations and SQL 2000.
We found that all things being equal SQL 2005 64-bit is roughly 20% faster in serving up content then SQL 2K. This would make sense with the OS at 64-bit and using all 8GB of ram. That should come as no surprise and I hope nobody seriously considers running SQL 2K in any type of large production implementation.
VMWare or any virtulization software seems to be OK for WSS with a remote sql instance but will not perform well for MOSS unless you allocate so many resources to those server instances that you might as well have purchased a server.
MOSS 64-bit vs 32-bit: 64-bit is clearly faster than 32-bit but I couldn’t tell if that was simply due to the OS using all 8GB of ram. MOSS loves memory. Installing 64-bit should have been required as in Exchange 2007.
F5’s Big-IP 6400’s are simply fantastic is performing web acceleration and load balancing for SharePoint. Our stress tools showed an average 20% decrease in time between first block received when using the web acceleration settings on the Big-IP.
Let me know what you are using or if you find contrary performance indicators!!
