Windows 2000 Advanced Server Cluster Storage

This article provides an overview of the shared storage architectures required to support cluster service failover in Microsoft Windows 2000 Advanced Server. In this operating system, Microsoft Cluster Service (MSCS) relies on a physical shared-disk topology coupled with a logical "shared-nothing" clustering software model. Below is an examination of the supported hardware interconnects, disk requirements, and arbitration mechanisms that allow seamless failover between cluster nodes.

Shared-Disk Hardware Architecture

Windows 2000 Advanced Server supports up to a two-node cluster configuration. For failover to occur, both server nodes must have concurrent physical access to the same external storage devices. This is achieved through two primary storage architectures:

  • Shared SCSI Buses: Multiple initiator configurations using high-voltage differential (HVD) or low-voltage differential (LVD) parallel SCSI buses. Both nodes connect to the same external SCSI enclosure, requiring specialized bus termination and unique SCSI IDs for each host adapter and storage target.
  • Fibre Channel Storage Area Networks (SANs): External storage connected via Fibre Channel Arbitrated Loop (FC-AL) or switched Fibre Channel fabrics. This architecture provides higher throughput, greater distance between nodes, and better scalability compared to parallel SCSI.

The "Shared-Nothing" Clustering Model

Although the physical storage hardware is shared, Windows 2000 Advanced Server operates on a "shared-nothing" software model. Under this architecture:

  • Only one node owns and accesses a given disk volume at any single time.
  • Disks are not concurrently read or written by both nodes simultaneously, preventing file system corruption.
  • During a failover event, the surviving node takes ownership of the storage resources previously managed by the failed node.

SCSI Reservations and Volume Management

To enforce single-node ownership, the Cluster Service uses SCSI reserve and release commands:

  • Bus-Level Arbitration: The active node places a SCSI reservation on the storage resource. If that node fails or stops responding to heartbeat signals, the secondary node issues a SCSI bus reset and asserts its own reservation to take control of the volume.
  • File System Requirements: All shared cluster disks must be formatted with the NTFS file system. Windows 2000 does not support FAT or FAT32 partitions on shared cluster storage.
  • Basic Disks Only: Shared disks must be configured as basic disks containing standard primary partitions. Dynamic disks and software-based fault-tolerant volumes (such as software RAID-5 or striped sets created in the OS) are not supported on shared cluster buses; hardware-level RAID controllers inside the external array must be used instead.

The Quorum Resource

A critical component of this architecture is the Quorum disk. Located on the shared storage bus, the Quorum is a dedicated NTFS physical disk resource that maintains the cluster configuration database (the cluster registry hive) and log files. It acts as the ultimate tie-breaker during network partitions ("split-brain" scenarios), ensuring that only the node able to reserve and write to the Quorum disk can form or maintain the active cluster.