VAXcluster Systems Quorum

This technical journal for VAXcluster System Management presents a collection of articles focusing on performance, disaster tolerance, network optimization, and revision levels within VAXcluster environments.

The first article, "Volume Shadowing for OpenVMS Version 5.5-2 Performance," compares Phase I (controller-based) and Phase II (host-based) implementations of Volume Shadowing, a RAID-1 product. It highlights that while Phase I prioritized fast copy/merge operations at the expense of user I/O, Phase II (pre-5.5-2) prioritized user I/O by making copy/merge operations subservient. Version 5.5-2 introduces significant enhancements to Phase II, notably "assisted" copy and merge operations. Assisted copies leverage the HSC controller for disk-to-disk data movement, reducing host CPU and interconnect bandwidth, and performing faster than Phase I for single operations. "Minimerge" operations, also introduced in 5.5-2, utilize controller memory write logs to quickly identify and correct disk differences after a node failure, making merge operations nearly invisible to normal system use and far superior to earlier methods.

The second article, "An Introduction to Disaster-Tolerant VAXcluster Systems," discusses the evolution of VAXcluster technology towards higher resource availability, including support for geographically distributed nodes via FDDI. Disaster tolerance is achieved through site redundancy, requiring duplication of critical hardware and software across two separate locations. Key redundancy requirements include sufficient CPU/memory, localized system disks, and shadowed critical data disks distributed between sites. Disaster recovery plans address loss of a site or communication between sites. The concept of a "quorum system," an inexpensive third-site system providing a tie-breaker vote, is introduced as a way to automate quorum restoration and avoid manual intervention in certain failure scenarios. The Operations Management Station (OMS) software is presented as a tool for managing multisite VAXcluster configurations, integrating them into a single logical datacenter.

The third article, "Understanding DECnet-VAX Phase IV Executor Pipeline Quota," addresses performance anomalies in LAN-based DECnet-VAX Phase IV environments, particularly affecting PATHWORKS and DECwindows Motif. These issues, characterized by inconsistent I/O rates and receiver overruns, are attributed to a misunderstanding of executor pipeline quota usage and buffer management, compounded by Ethernet adapter variations. The recommended solution is to tune the DECnet-VAX executor pipeline quota; setting it to 4032 (seven transmit buffers) is advised for most LAN-based VAX systems, while 1728 (three buffers) is better for slower PCs/CPUs to prevent receiver overruns and improve efficiency. The article details how pipeline quota affects transmit window size, describes different buffer layers, and explains the impact of Ethernet adapter hardware on network behavior and collision rates.

Finally, "R1 and S1 Revision Management Level" lists the minimum acceptable revision levels for various VAXcluster components, including specific VAX systems, adapters, disk drives, tape subsystems, and software. It notes that Revision Management Level R1 supports VMS Version 5.4-3 and numerous hardware components, while S1 supports VMS Version 5.5 and other hardware. The article advises upgrading VAXcluster systems to the latest revision for reliable operation.