We performed two test cycles per product, one each with Exchange 2000 running on Windows 2000 Advanced Server and one running with Exchange 2003 on Windows 2003 Enterprise Server.

We used an Exchange installation of 500 user accounts, an Active Directory containing the same number of users and a mail store of 30G bytes. We partitioned the servers into 500 total users at two sites, the first site having 300 and the other 200. We also created a public shared folder store within Exchange with 10G bytes of data.

Site A comprised two Compaq DL360G3 servers (twin 3.0-GHz P4 CPUs; 1G byte DRAM) connected by Emulex 2FC controllers into a storage-area network (two independent JMR Fortra non-cached RAID 5 drives) via a Brocade Silkworm 3800 2FC switch. This network was connected via 10Base-T (768K bit/sec to simulate a WAN connection speed) link to an identical network at Site B. The exception to this was the test setup for LeftHand, which uses its own storage.

The second hotsite/standby site had a primary and secondary server available for each product. We connected the two sites via a Point-to-Point Tunneling Protocol VPN. How the products were installed to pick up or have concurrent availability varied widely: See the individual product test experiences for specifics.

One of the servers was an Active Directory Global Catalog Server with 300 users, and the other had 200 users as a partition of the forest of users. The primary site had, therefore, two servers, not identical in storage or users. We used VMWare GSX (from EMC) running on Micron servers (twin 3.2-GHz P4s with 2G bytes of DRAM) to simulate 24 users making transactions in a pattern. Once the pattern started and at exactly the same place/time during each test run, we pulled the plugs on the server. This process demonstrated whether the messages were being replicated in near-real time or whether the message/block/I/O replication method of the application used snapshots every so often, and would miss the messages that were in the Exchange servers - but not sent - correctly.