VMScluster

VAXcluster / VMScluster
Company type	Computer clustering technology
Industry	Computer hardware / Software
Founded	May 1983; 43 years ago
Founder	Digital Equipment Corporation (DEC)
Products	VAXcluster, VMScluster clustering technology
Parent	Digital Equipment Corporation

A VMScluster, initially known as a VAXcluster, refers to a computer cluster composed of systems running the OpenVMS operating system. Unlike tightly coupled multiprocessors that operate under a single OS instance, a VMScluster employs loosely coupled architecture: each node runs its own OpenVMS environment, yet disk storage, the lock manager, and security mechanisms function across the entire cluster, delivering a single system image. Nodes can be added or removed dynamically without interrupting cluster operations. To enhance reliability, VMSclusters support dual-ported disks that can be accessed concurrently by two machines or controllers.

Initial release

In May 1983, Digital Equipment Corporation (DEC) introduced VAXclusters. Implementing clustering required both specialized communication hardware and extensive modifications to the operating system, which were developed in tandem with the hardware components. The first VAXcluster support appeared in VAX/VMS version 4.0 (1984), initially limited to DEC’s proprietary Computer Interconnect (CI).

Each cluster was structured around a star coupler, which linked every node and storage device through one or two pairs of CI cables. Each cable pair delivered 70 megabits per second—a significant speed at the time—with redundancy doubling this to 140 megabits per second to ensure fault tolerance. Star couplers also incorporated duplicated wiring paths to maintain availability.

Nodes connected to CI cables via CI Ports, which managed packet transmission autonomously, minimizing CPU involvement. The CPUs simply queued data structures for sending; incoming packets were queued similarly. Performance tests demonstrated that a VAX-11/780 (a 1-MIPS system) could handle 3000 messages per second. The related Mass Storage Control Protocol (MSCP) provided similar throughput for mass storage access and allowed MSCP packets to be transmitted over CI, enabling remote storage utilization.

VAXclustering was among the earliest commercially viable clustering technologies—alongside solutions from AT&T, Tandem Computers, and Stratus Computers—and it played a key role in driving VAX system sales growth.

Later developments

Initially, DEC’s MicroVAX minicomputer did not support VAXclustering.^[1] In 1986, DEC expanded VAXclustering capabilities to operate over Ethernet rather than relying solely on specialized hardware. These Local Area VAXclusters (LAVc) lacked the high-availability features of CI-based systems but offered an attractive upgrade path for lower-end minicomputer customers. LAVc also introduced the concept of diskless satellite nodes, which could boot via the network using a system disk hosted on a bootnode.

Subsequent OpenVMS releases (version 5.0 and beyond) enabled "mixed interconnect" VAXclusters that combined CI and Ethernet connections, as well as clustering over DSSI, SCSI, and FDDI. With improvements in wide area networking technology, clustering expanded to include satellite links and long-distance terrestrial connections, facilitating disaster-tolerant clusters capable of surviving infrastructure outages by spanning multiple geographic regions.

The deployment of terminal servers utilizing the LAT protocol simplified terminal access across cluster nodes through Ethernet, allowing transparent user terminal fail over within the cluster environment.

Clustering support over TCP/IP networks was introduced in OpenVMS 8.4 (2010). With the widespread use of Gigabit Ethernet and emerging 10 Gigabit Ethernet, standard network infrastructure became sufficient for implementing VMSclustering without specialized hardware.

Features

OpenVMS accommodates clusters with up to 96 nodes and supports heterogeneous environments that integrate VAX, Alpha, and Itanium architectures. While configurations featuring as many as 150 nodes and triple-architecture combinations have been demonstrated unofficially, these setups are not officially endorsed.

Distinct from many clustering technologies, VMScluster delivers seamless, fully distributed read-write operations with record-level locking. This functionality permits concurrent access by multiple nodes to identical disks or files while restricting locks to individual records—such as entries within a database—facilitating highly available and fault-tolerant database servers.

Clusters are capable of extending over considerable distances, enabling connections between nodes situated across multiple campuses or even different cities.

Volume shadowing at the host level duplicates data across several controllers and hosts, providing robust disaster recovery through fault-tolerant configurations.

Developers can utilize the distributed lock manager (DLM) extensively, coordinating various resources and processes cluster-wide beyond the scope of file-level locking.

The system supports rolling upgrades alongside multiple system disks, allowing for online maintenance and phased software updates without interrupting application operations or data accessibility.

Cluster systems often maintain continuous operation for years; the longest documented uptime surpasses sixteen years.

Initial release

Later developments

Features

Further reading