NEC Upgrades Windows Fault Tolerant Servers

by Timothy Prickett Morgan

NEC Solutions America, the U.S. unit of Japanese server maker NEC, has announced an upgrade to its fault tolerant FT Server Windows server line. The upgrade provides more computing power to customers as well as better fault tolerance though the inclusion of features from NEC's ExpressCluster high availability software. While the extra oomph is welcome, customers will probably find the new--and free--ExpressCluster features more interesting.

The new Express5800/320Lc is a two-processor server that uses 32-bit Xeon DP processors running at 2.8 GHz. The existing Express5800/320Lb server, which uses 2.4 GHz Xeon DPs, is still available for Windows shops who want that machine, says Brad Lightner, director of product marketing at NEC's American unit. Both machines offer one or two physical servers that are mirrored at the hardware level using fault tolerant clustering technologies developed by Stratus Technologies.

The way fault tolerant machines work, NEC and Stratus double up on the core server components and do the same work twice; if one component in the clustered systems fails, another machine has already done the exact same work and downtime is minimized. NEC launched its Express5800/ft line for Windows 2000, also based on Stratus intellectual property, back in May 2001. In September 2004, NEC launched the Express5800/340 Hb-R, a machine with up to four logical processors (that's eight physical processors) that runs Windows Server 2003 (and even includes a license to it). NEC adds proprietary hardware and software (mostly for remote systems management) to the Stratus technology that it licenses to create this fault tolerant Windows box; it is not merely a re-branding of an existing Stratus machine.

The Express5800/320Lc is a 4U rack-mounted server that supports up to two logical processors (that's four physical processors). It also includes redundant I/O systems and can span up to 6 GB logical (12 GB physical) of main memory. Each half of the server has one 100 Mbit and one Gigabit Ethernet port. Each I/O module has support for three disk drives and a CD-ROM. NEC is supporting 10K RPM drives in 36 GB, 73 GB, and 146 GB capacities and 15K drives in 18 GB and 73 GB capacities. Customers with the need to store more data can hook up NEC's S2800 Fibre Channel RAID arrays to the FT Server line, boosting storage capacity to 27 TB and write cache to 8 GB per dual controller. Each physical motherboard in the system has three PCI slots.

Lightner says that NEC wants an edge in the Windows market and it wants to address some high availability concerns that even fault tolerant server hardware cannot address. So NEC has decided to bundle in a subset of its ExpressCluster high availability software on its Windows-based and bare-bones servers. (You can buy an FT Server without Windows, if you want to put Linux or another operating system on the box.) That subset of ExpressCluster is called Self-Recovery Edition, and it includes software monitoring tools and clustering software that will allow the FT Servers to cope with the software errors that account for about 20 percent of server crashes out there in the real world. These software errors cannot be handled by a fault tolerant server, which is designed to keep the failure of critical hardware components from keeping an application from coming down. To get true high availability, you have to do more than keep the hardware running.

ExpressCluster SRE was originally designed for disaster recovery (meaning to replicate data and applications in a remote site in the event that a disaster took out the primary data center). With the SRE monitoring software running, the FT Server can monitor applications and the Windows operating system; if a software crash is imminent in the application, SRE can automatically restart the application before it fails. If that doesn't correct the problem, then SRE can restart the Windows operating system and the applications. SRE does this without any involvement of a system operator. With the fault tolerant server plus SRE option, which is valued at $3,000 to $5,000, customers do not have to set up high availability clusters of Windows servers to get the benefits of an HA cluster.

The Express5800/320Lc began shipping last week. A base machine with dual logical processors (4 physical), 512 MB of logical main memory, a Windows 2003 Enterprise Edition license, and the ExpressCluster SRE software sells for $24,999. NEC has dropped the price of the Express5800/320Lb server, which uses the 2.4 GHz Xeon DP chips, to $22,499, and Lightner says that the company will be repositioning this machine as a Linux box, and is working on a special fault tolerant version of Linux with Stratus based on the Linux 2.4 kernel for the box.