Marathon Technologies everRun HA

Server fault tolerance is highly desirable for business-critical applications, but can be too costly for the smaller business. However, those that are prepared to accept fault tolerance at the component rather the server level may find Marathon's everRun HA is just what they've been looking for.

At the hardware level, there are many similarities between HA and Marathon's everRun FT, as they both require a pair of servers. Called CoServers, these present a single virtual server to the network and use disk mirroring to synchronise both storage systems. The big difference is that, as FT performs lock-stepping across them, the servers need to be identical twins. Not so with HA, as only one CoServer actually manages the virtual server. This is coined the Active server, and if any component fails then all I/O is redirected to the equivalent hardware in the Ready server. However, in the event of a complete failure of the Active server, there'll be a brief interruption in services, as HA must migrate the virtual environment across to the Ready server and reboot it. Consequently, transactions that occurred during this phase will probably be lost.

HA supports up to four logical processors with Windows Server 2003 (WS2003) Standard and up to eight with the Enterprise version. Standard installations see a limit of 4GB of memory, but Enterprise's limit is 32GB.

Installation is virtually identical to FT. Each server requires WS2003 with SP1 loaded first, followed by the HA software. We used a pair of HP ProLiant ML370 G4 servers with dual 3.2GHz Xeon CPUs. During this phase, you decide how the servers are to be connected to allow for disk mirroring, I/O redirection and presenting services to the network over a virtual connection. You use the HA management interface to create a virtual boot disk and then load Windows on it. You must use a slipstreamed copy of WS2003 that has SP1 included; this can't be applied later on.

Management options abound. The HA Manager utility allows the virtual server to be accessed locally from within the virtual server desktop, from either CoServer, via a Remote Desktop link or from a separate Windows client on the optional management LAN. The Device Redirector tool is used to create virtual volumes and these are mirrored automatically. There's a bonus to using identical servers, as you can migrate the virtual server from Active to Ready systems without bringing down the system. We tested this and watched the virtual server move over in only eight seconds. With both servers in action, we powered the Active down to simulate a complete failure. HA immediately shifted across to the Ready server and, after a reboot, was up and running in under 50 seconds.

Hardware solutions at this level of the market are far too expensive and can lock you into a proprietary platform. The sacrifice is full server fault tolerance, but everRun HA does offer a very cost-effective alternative.

Author: Dave Mitchell