It's hard to kill Stratus Technologies' ftServer 5700. We tried. We tested the highly fault-tolerant system in a number of very hostile ways, and at worst it took about 6 minutes to recover to full synchronization -- while continuing to run. The lock-step synchronization of what amounts to two servers in a 4U frame wasn't terribly difficult to manage. We found the software to be potentially dangerous, however, because of the Spartan instrumentation that Stratus uses.
Where it counts -- in not losing transactions and staying alive -- the ftServer 5700 couldn't be made to fail.
Stratus offers three groups of servers -- the ftServer 5700 for Windows Server, and versions for Linux and Stratus' Multics-like VOS operating system. The ftServer 5700 is two 1U server systems in one package. To the outside world, it places its network driver ahead of others, to control network data flow so that one IP address is represented for all the Ethernet connections going to the machine.
A Stratus Ethernet interface and driver must be installed and set, but we found this simple to do, although in testing they were slow. After the servers are installed, the death or crippling of either computer system has no effect -- the ftServer 5700 continues to run.
This durability has benefits and detractions. As a single unit, the ftServer 5700 is a very highly reliable system. Its dual-home power supplies and dual-everything-else (including the operating system and applications) make it resistant to hardware failures, hard-drive failures especially. The downside is that abnormal ambient conditions -- such as heat, condensing humidity, a hurricane -- could kill both systems. Barring such events, the ftServer 5700 is highly available.
The setup conundrum
The Stratus arrived fully configured, with an operating system license installed. One instance of Windows 2003 Server was included. Although the Stratus has two separate systems, in terms of Windows license issues it's seen as one. We put the systems into lock-step mode, creating parallel partitions on each one, then forced an initial lock-step synchronization.
In our tests, we couldn't fail the ftServer 5700, but Stratus called us immediately after each outage. We discovered we could turn this feature off, but you have to tell Stratus, rather than just select an "It's OK, we're just fooling around" option.
We installed Exchange 2003 Enterprise and populated it with our users, then started pounding it with 125MB messages, increasing the number by one each time. We pulled the AC power from a unit arbitrarily, and the system continued to function despite minor hesitation. To the outside world, it kept ticking, sessions weren't destroyed, and Outlook or our SMTP mail transfer agent didn't require resetting the connection; this is unlike many failover products that disconnect then reconnect without having to abandon sessions or transactions.
Even though we couldn't kill the ftServer 5700, we found downsides. First and most glaring was that we couldn't get its Ethernet transfer rate to climb above 9Mbps in any mode. It went slower with certain types of connections, crawling at 1.4Mbps with an SSH secure copy from a Unix box.
The Stratus administration and management application is a single Microsoft Management Console snap-in that we found a bit dangerous if the disk arrays become uncoupled. Several procedures begged for wizards, as well. But the controls were effective, and the software (except for the Ethernet slowness) was fast. If you pay US$33,000 for the system, it's likely you'll read the directions and avoid a few of the possible missteps that could bring the server synchronicity down. If you don't, you can kill the servers fairly easily and accidentally.