NEC beefs up, slims down its fault-tolerant server line
- By John Breeden II
- Jul 30, 2003
There are times when downtime is not an option.
Many agencies have turned to clustering technology for redundancy to keep their systems running. Clustered servers share their workloads and, in case of failure, can take over one another's tasks.
Cluster hardware is relatively expensive, however, and difficult to administer. Not every application is cluster-aware, automatically knowing how to behave with multiple CPUs. Agencies that choose clustering often have to rewrite at least some of their code, and that's beyond the technical ability of most network administrators.
Furthermore, a cluster rollover is not always instantaneous after a failure.
NEC Solutions America Inc. has come up with an alternative: the fault-tolerant server. It's two identical systems running in tandem, while the operating system'in this case, Microsoft Windows 2000 Advanced Server'sees only a single system. The duplicate components automatically compensate for any point of failure. As a result, no downtime.
And as a bonus, the administrator needs to license only a single copy of the operating system and a standard copy of each program in use'not a cluster version.
When the GCN Lab tested the first NEC f/t server model, we were impressed by its reliability, but we noted a couple of disadvantages: processor speed and unit size.
The new Express5800/320Lb takes care of both objections.
Sizewise, NEC has embraced the blade design. A fully redundant server now stands only four rackmount units high'less than 7 inches. If you prefer, you can still buy the larger pedestal platform.Faster, better
Speed and performance have improved. The earlier 800-MHz processors have given way to dual 2.4-GHz Intel Xeons. They performed well in our benchmark testing. Remember, however, that the operating system thinks the dual CPUs are just one, so you have only single processor power.
At 800 MHz, the earlier NEC f/t was a little slow, but most applications should have no problem with the current 2.4-GHz CPUs.
We also liked the 512M of RAM, 36.4G hard drive and PCI module with 10/100/1,000-Mbps integrated network card. Again, each of these components was replicated to eliminate a single point of failure.
In our tests for fault tolerance, the 320Lb performed perfectly every time. We removed a disk drive during a write and even unplugged one of the two power supplies. Except for issuing a power warning, the OS didn't even notice.
The only time the server hiccuped for a second was when we reinstalled a disconnected module. The server paused to reintegrate and synchronize the module with its mirrored module. We noticed the pause only while running a movie or other processor- or disk-intensive operation. And it was only a momentary pause, not a crash.
The Express5800/320Lb adds teeth to NEC's fault-tolerant lineup. It maintains the advantages of the older, slower f/t system while significantly raising speed and changing to a blade design.
It's perfect for environments where even a few hours of downtime are not an option.