To compare performance of the 64-bit and 32-bit versions of Windows Server 2003 Enterprise Edition, I used the standard DirectoryMark benchmark. The tests, which involved 10 DirMark clients running two threads each, were performed on an Intel-supplied Itanium 2 machine (four 1GHz processors, 3MB Level 2 cache, 32GB of RAM) and a Dell 6650 Xeon server (four 1.5GHz processors, 3MB Level 2 cache, 8GB of RAM). The storage array for both systems was an IBM TotalStorage EXP 600 and FAStT 700 storage server, providing a total of 1.98TB of disk space (42 x 73.4GB 10,000 rpm drives).
I loaded five million objects into Active Directory on both systems, and performed the standard, timed load of 10 minutes. The tests consisted of DirectoryMark’s Global Address Lookups and E-Mail Messaging Simulation. For more information on the exact queries performed, consult the DirectoryMark Web site.
I tested the 64-bit versions of SQL Server 2000 and Oracle9i, both on Windows Server 2003, under the same conditions. I loaded 200GB of data onto a storage array configured with RAID 5 sitting on the Itanium 2 machine, and placed a load of general busy work on the server by performing very large joins and temp table operations in infinite loops. I then simulated increments of users up to 500, with each user doing 15 separate OLTP (online transaction processing) operations. Each user performed a series of selects, inserts, and updates (some quite large). I controlled excessive caching by clearing the memory periodically. My goal was to see how the system performed under great stress, so I kept indexes to a bare minimum.