Cluster Server to Scale Beyond Two Nodes

Microsoft Corp.’s Cluster Server (MSCS) receives negative press for its inability to scale beyond two node failover clustering. Some companies and analysts say the inability to cluster more than two nodes is a weakness. Others, citing the more polished clustering capabilities of Unix, maintain that even if Windows NT could cluster beyond two nodes, very few IT departments would do so anyway.

Criticisms aside, Microsoft plans to add more nodes to the capabilities of MSCS, and enhance the product through its own development efforts and forging partnerships.

One major shortcoming of MSCS is its "shared nothing" approach, which means that while several systems in the cluster may have access to a device or resource, each resource is effectively owned and managed by a single system at a time. This means that MSCS enhances availability, but has limited ability to enhance performance.

A nice complement to MSCS is Windows Load Balancing Service (WLBS). This product is based on Convoy Clustering Software, which Microsoft recently acquired.

In an environment with many user connections, such as high volume Web serving, WLBS offers significant performance and reliability enhancements over traditional approaches.

"WLBS extends our clustering technology in some very important ways," says Karan Khana, lead product manager with Windows NT Server marketing. "If you visualize an n-tier application, the Convoy technology provides load balancing and high availability for the first tier -- the user interface."

This is TCP/IP load balancing -- so several workloads can benefit from the technology, including the load balancing of IIS sessions, and PPTP and VPN sessions. Up to 32 Windows NT servers can be formed into a Convoy cluster.

MSCS will be improved with the release of Windows 2000 Advanced Server, and later with Windows 2000 Datacenter Server. Key enhancements will include usability, support for cluster failover and upgrade support.

Improved usability consists of wizards for application and virtual server setup, and easier user interfaces for setup and management. Extensive support for cluster failover of system services includes all existing services, such as IIS, file, print and MSMQ, and new support for DHCP, WINS and DfS. Rolling upgrade support allows service packs, hot fixes, and software updates to be applied to individual cluster nodes without taking the cluster as a whole offline. Rolling upgrade support for clusters is now available with NT 4’s service pack 4.

Other enhancements featured are support for WBEM and MMC, and integration with Active Directory, as well as NIC failure detection and failover. Windows 2000 Advanced and Datacenter servers also include COM+ Load Balancing.

Although the improvements coming up in the next few months are nice, most scalability action will take place in the post-Windows 2000 timeframe, when Microsoft anticipates adding support for greater than two node clustering. Khana says work is progressing on this front, but Microsoft will not commit to a specific number of nodes or a time of release.

Microsoft and Compaq Computer Corp. are jointly developing advanced clustering, OLTP scalability and robustness technologies. The agreement covers six clustering-related technologies that are only available on Digital OpenVMS, the Tandem NonStop Kernel, or Digital Unix.

Down the road, Microsoft plans to deliver these capabilities to Windows NT. The enhancements will include clustered transactional and recovery services, remote mirroring, clustered file system infrastructure and management for concurrent read/write access to data files. Additionally, Microsoft would like to add to MSCS data and file partitioning infrastructure, advanced software tools for high availability systems operation, remote system management and system health detection services.

Although Microsoft’s clustering technology has a good start, MSCS still needs some development to cluster as well as Unix. WLBS is an innovative complement, and the enhancements that will come with Windows 2000 Advanced and Datacenter servers represent incremental improvements. Ultimately, if Microsoft is serious about the enterprise, it needs to outgrow its shared-nothing approach to clusters and adapt and enhance the technology Compaq agreed to jointly develop.