IBM Demos 8-node Clustering on NT
- By Scott Bekker
DALLAS -- Clustering capabilities on Windows NT got a turbo boost at Tech Ed when IBM Corp. first demonstrated a solution that allows up to 8-node, any-to-any failover clustering technology built atop Microsoft Cluster Services (MSCS).
A demonstration of a 5-node IBM/NT cluster during Microsoft Corp. Windows 2000 chief Brian Valentine’s TechEd keynote Friday supported Microsoft’s theme that the Windows NT/Windows 2000 platform is gaining momentum in availability and reliability.
The IBM solution, code-named Cornhusker, will be formally announced and released this summer as a hardware, software and services package. The formal name is Netfinity Availability Extensions for Microsoft Cluster Services, and it represents a large jump over the 2-node capability of MSCS, still also known by its code-name Wolfpack.
"We are the only ones who have actually extended what Microsoft has [in MSCS]," says Roger Hellman, worldwide product manager, Netfinity, Personal Systems Group.
IBM’s willingness to work with Microsoft and preserve the MSCS API prompted Microsoft to cooperate fully, says Michel Gambier, Microsoft’s product manager for Windows 2000 Enterprise Marketing. "Those extensions by IBM do not break any applications," Gambier says. "That was really one of the keys for us to work together on it."
Use of the MSCS API means Cornhusker will work on Windows NT 4.0 Enterprise Edition, Windows 2000 and clusters that involve mixes of both server types, says Randy Groves, IBM’s vice president of Netfinity development.
Groves contends that while Microsoft’s 2-node clusters probably satisfy 70 percent to 80 percent of the market for failover clustering, IBM’s extensions bring the capabilities of the cluster-aware database, mail, groupware, Web-serving and file-and-print applications written for Windows NT to enterprises with more extensive high-availability needs.
For Friday’s demonstration, IBM put five Netfinity servers on the stage running file-and-print with 1,000 simulated users. First, power was taken off one of the servers, and it failed over to another of the five. Then the dual-loaded server was taken off line, along with two of the other servers. The fifth server picked up the workloads of all four failed servers.
According to IBM officials, what the demonstration means is that a company that wanted high availability in the past for 4 servers with MSCS needed 8 servers: 4 production servers tied in 4 Wolfpack clusters to 4 fully available backups. With Cornhusker, that same scenario could be satisfied with 5 servers: 4 production servers and 1 server fully available as a backup.
"You get the same level of availability at 5/8ths, roughly, the cost," Hellman says. "That’s one of the reasons that people are interested in greater than 2-nodes."
In a room elsewhere in the Dallas Convention Center, IBM pushed the limits of the new offering with a demonstration of a cluster of its high-end Netfinity 7000 M10 servers running a single IBM DB2 UDB database query across 8 nodes.
For its part, Microsoft is also becoming more vocal about its own greater-than-2-node clustering efforts. Valentine said Friday that Windows 2000 Datacenter Server, a high-end version of Windows 2000 server expected to ship several months after the rest of the product, will include support for 4-node failover clustering. He also promised that eventually, "We’ll go beyond 4-way. The architecture supports well beyond 4-way."
In other availability announcements, Valentine announced two Web resources for IT professionals, both to be found at www.microsoft.com/ntserver. One is a high-availability deployment guide that documents the experience and best practices of nine customers. The other is an Uptime Tool for quantifying data about how long an NT system has been online and for reporting on all system reboots. -- Scott Bekker
Scott Bekker is editor in chief of Redmond Channel Partner magazine.