Updated: 11 December 1998

Guidelines for OpenVMS Cluster Configurations

This chapter provides information to help you select systems for your OpenVMS Cluster to satisfy your business and application requirements.

3.1 Alpha and VAX Architectures

An OpenVMS Cluster can include systems running OpenVMS Alpha, OpenVMS VAX, or both. Compaq provides a full range of systems for both the Alpha and VAX architectures.

• OpenVMS Alpha operating system
  Based on a 64-bit RISC (reduced instruction set computing) architecture, OpenVMS Alpha provides industry-leading price/performance benefits with standard I/O subsystems for flexibility and expansion.
• OpenVMS VAX operating system
  Based on a 32-bit CISC (complex instruction set computing) architecture, OpenVMS VAX provides high CISC performance and a rich, powerful instruction set. OpenVMS VAX also supports a wide variety of standard I/O subsystems.

Alpha and VAX systems span a range of computing environments, including:

• Personal computers (PCs)
• Workstations
• Departmental systems
• Enterprise systems
• Data center systems

Your choice of systems depends on your business, your application needs, and your budget. With a high-level understanding of systems and their characteristics, you can make better choices.

Table 3-1 is a comparison of recently shipped OpenVMS Cluster systems. While laptop and personal computers can be configured in an OpenVMS Cluster as client satellites, they are not discussed extensively in this manual. For more information about configuring PCs and laptops, see the PATHWORKS Version 6.0 for DOS and Windows: Installation and Configuration Guide

Table 3-1 System Types

System Type	Useful for	Examples
Workstations	Users who require their own systems with high processor performance. Examples include users running mechanical computer-aided design, scientific analysis, and data-reduction and display applications. Workstations offer the following features: Lower cost than departmental and data center systems Small footprint Useful for modeling and imaging 2D and 3D graphics capabilities	DIGITAL Personal Workstation DPWau series AlphaStation 200 AlphaStation 500 AlphaStation 600 VAXstation 4000 MicroVAX 3100
Departmental systems	Midrange office computing. Departmental systems offer the following capabilities: High processor and I/O performance Supports a moderate number of users, client PCs, and workstations	AlphaServer 400 AlphaServer 1000 AlphaServer 1000A AlphaServer 1200 AlphaServer 2000 AlphaServer 2100 AlphaServer 4100 VAX 4000
Data center systems	Large-capacity configurations and highly available technical and commercial applications. Data center systems have a high degree of expandability and flexibility and offer the following features: Highest CPU and I/O performance Ability to support thousands of terminal users, hundreds of PC clients, and up to 95 workstations	AlphaServer 8400 AlphaServer 8200 VAX 7800

3.4 Scalability Considerations

When you choose a system based on scalability, consider the following:

• Maximum processor capacity
• Maximum memory capacity
• Maximum storage capacity

The OpenVMS environment offers a wide range of alternative ways for growing and expanding processing capabilities of a data center, including the following:

• Many Alpha and VAX systems can be expanded to include additional memory, processors, or I/O subsystems.
• You can add systems to your OpenVMS Cluster at any time to support increased work load. The vast range of systems, from small workstation to high-end symmetric multiprocessing (SMP) systems, can interconnect and be reconfigured easily to meet growing needs.
• You can add storage to your OpenVMS Cluster system by increasing the quantity and speed of disks, CD-ROM devices, and tapes.
  Reference: For more information about storage devices, see Chapter 5.

Reference: For more information about scalability, see Chapter 10.

3.5 Availability Considerations

An OpenVMS Cluster system is a highly integrated environment in which multiple systems share access to resources. This resource sharing increases the availability of services and data. OpenVMS Cluster systems also offer failover mechanisms that are transparent and automatic, and require little intervention by the system manager or the user.

Reference: See Chapter 8 and Chapter 9 for more information about these failover mechanisms and about availability.

3.6 Performance Considerations

The following factors affect the performance of systems:

• Applications and their performance requirements
• The number of users that the system must support
• The type of storage subsystem that you require

With these requirements in mind, compare processor performance, I/O throughput, memory capacity, and disk capacity in the Alpha and VAX specifications that follow.

3.7 System Specifications

The DIGITAL Systems and Options Catalog provides ordering and configuring information for Intel, Alpha, and VAX workstations and servers. It also contains detailed information about storage devices, printers, and network application support.

To access the most recent DIGITAL Systems and Options Catalog on the World Wide Web, use the following URL:

http://www.digital.com:80/info/soc/

An interconnect is a hardware connection between OpenVMS Cluster nodes over which the nodes can communicate. This chapter contains information about the following interconnects and how they are used in OpenVMS Clusters:

• Fibre Channel (storage only)
• MEMORY CHANNEL (node-to-node only)
• SCSI (Small Computer Systems Interface) (storage only)
• CI (computer interconnect)
• DSSI (Digital Storage Systems Interconnect)
• Ethernet
• FDDI (Fiber Distributed Data Interface)

The software that enables OpenVMS Cluster systems to communicate over an interconnect is the System Communications Services (SCS).

4.1 Characteristics

The six interconnects described in this chapter share some general characteristics. Table 4-1 describes these characteristics.

Table 4-1 Interconnect Characteristics

Characteristic	Description
Throughput	The quantity of data transferred across the interconnect. Some interconnects require more processor overhead than others. For example, Ethernet and FDDI interconnects require more processor overhead than do CI or DSSI. Larger packet sizes allow higher data-transfer rates (throughput) than do smaller packet sizes.
Cable length	Interconnects range in length from 3 m to 40 km.
Maximum number of nodes	The number of nodes that can connect to an interconnect varies among interconnect types. Be sure to consider this when configuring your OpenVMS Cluster system.
Supported systems and storage	Each OpenVMS Cluster node and storage subsystem requires an adapter to connect the internal system bus to the interconnect. First consider the storage and processor I/O performance, then the adapter performance, when choosing an interconnect type.

4.2 Comparison of Interconnect Types

Table 4-2 shows key statistics for a variety of interconnects.

Table 4-2 Comparison of Interconnect Types

Attribute	CI	DSSI	FDDI	SCSI	MEMORY CHANNEL	Ethernet	Fibre Channel
Maximum throughput (Mb/s)	140	32	100	160	800	10	1000
Hardware-assisted data link 1	Yes	Yes	No	No	No	No	No
Connection to storage	Direct and MSCP served	Direct and MSCP served	MSCP served	Direct and MSCP served	MSCP served	MSCP served	Direct and MSCP served
Topology	Radial coaxial cable	Bus	Dual ring of trees	Bus or radial to a hub	Radial copper cable	Linear coaxial cable	Radial to a switch
Maximum nodes	32 2	8 3	96 4	8--16 5	4	96 4	8 7
Maximum length	45 m	6 m 6	40 km	25 m	3 m	2800 m	400 m

You can use multiple interconnects to achieve the following benefits:

• Failover
  If one interconnect or adapter fails, the node communications automatically move to another interconnect.
• MSCP server load balancing
  In a multiple MSCP server configuration, an OpenVMS Cluster performs load balancing to automatically choose the best path. This reduces the chances that a single adapter could cause an I/O bottleneck. Depending on your configuration, multiple paths from one node to another node may transfer more information than would a single path.
  Reference: See Section 10.7.3 for an example of dynamic MSCP load balancing.

A mixed interconnect is a combination of two or more different types of interconnects in an OpenVMS Cluster system. You can use mixed interconnects to combine the advantages of each type and to expand your OpenVMS Cluster system. For example, an Ethernet cluster that requires more storage can expand with the addition of CI, DSSI, or SCSI connections.

4.5 Interconnects Supported by Alpha and VAX Systems

Table 4-3 shows the OpenVMS Cluster interconnects supported by Alpha and VAX systems. You can also refer to the most recent OpenVMS Cluster SPD to see the latest information on supported interconnects.

Table 4-3 Cluster Interconnects Supported by Systems

Systems	CI	DSSI	SCSI	FDDI	Ethernet	MEMORY CHANNEL	Fibre Channel
AlphaServer 8400, 8200	X	X	X	X 1	X	X	X
AlphaServer 4100, 2100, 2000	X	X	X	X 1	X 1	X	X 2
AlphaServer 1000		X	X	X	X 1	X
AlphaServer 400		X	X	X	X 1
AlphaStation series			X	X	X 1
DEC 7000/10000	X	X		X 1	X
DEC 4000		X		X	X 1
DEC 3000			X	X 1	X 1
DEC 2000				X	X 1
VAX 6000/7000/10000	X	X		X	X
VAX 4000, MicroVAX 3100		X		X	X 1
VAXstation 4000				X	X 1

As Table 4-3 shows, OpenVMS Clusters support a wide range of interconnects: CI, DSSI, SCSI, FDDI, Ethernet, and MEMORY CHANNEL. This power and flexibility means that almost anything will work well. The most important factor to consider is how much I/O you need, as explained in Chapter 2.

In most cases, the I/O requirements will be less than the capabilities of any one OpenVMS Cluster interconnect. Ensure that you have a reasonable surplus I/O capacity, then choose your interconnects based on other needed features.

4.6 Fibre Channel Interconnect

Fibre Channel is a high-performance ANSI standard network and storage interconnect for PCI-based Alpha systems. It is a full-duplex serial interconnect and can simultaneously transmit and receive 100 megabytes per second. For the initial release, Fibre Channel will support simultaneous access of SCSI storage by multiple nodes connected to a Fibre Channel switch. A second type of interconnect is needed for node-to-node communications. Node-to-node communications over Fibre Channel is planned for a future release.

For multihost access to Fibre Channel storage, the following components are required:

• Fibre Channel host adapter (KGPSA)
• Multimode fiber-optic cable (BNGBX-nn), where nn represents distance in meters
• Fibre Channel switch (DSGGA)
• Storage devices that are supported in a multihost configuration (HSG80)

The Fibre Channel interconnect offers the following advantages:

• High-speed transmission, 1.06 Gb/s
• Scalable configuration to support department to enterprise configurations. The first release of Fibre Channel supports up to four OpenVMS Cluster systems, up to four Fibre Channel switches, and up to four dual HSG storage controllers.
• Long-distance interconnects
  The first release of Fibre Channel supports multimode fiber at 400 meters per link. Longer distances are planned.
• High availability
  Multipath support is available. As many as four Fibre Channel interconnects can be connected to each OpenVMS Cluster system.

The Fibre Channel interconnect transmits up to 1.06 Gb/s. It is a full-duplex serial interconnect that can simultaneously transmit and receive 100 MB/s.

4.6.3 Supported Adapter

The Fibre Channel adapter, the KGPSA, connects to the PCI bus.

Reference: For complete information about each adapter's features and order numbers, see the DIGITAL Systems and Options Catalog.

To access the most recent DIGITAL Systems and Options Catalog on the World Wide Web, use the following URL:

http://www.digital.com:80/info/soc/

4.7 MEMORY CHANNEL Interconnect

MEMORY CHANNEL is a high-performance cluster interconnect technology for PCI-based Alpha systems. With the benefits of very low latency, high bandwidth, and direct memory access, MEMORY CHANNEL complements and extends the unique ability of OpenVMS Clusters to work as a single, virtual system.

Three hardware components are required by a node to support a MEMORY CHANNEL connection:

• A PCI-to-MEMORY CHANNEL adapter
• A link cable (3 m or 10 feet long)
• A port in a MEMORY CHANNEL hub (except for a two-node configuration in which the cable connects just two PCI adapters)

A MEMORY CHANNEL hub is a PC size unit that provides a connection among systems. MEMORY CHANNEL can support up to four Alpha nodes per hub. You can configure systems with two MEMORY CHANNEL adapters in order to provide failover in case an adapter fails. Each adapter must be connected to a different hub.

A MEMORY CHANNEL hub is not required in clusters that comprise only two nodes. In a two-node configuration, one PCI adapter is configured, using module jumpers, as a virtual hub.

4.7.1 Advantages

MEMORY CHANNEL technology provides the following features:

• Offers excellent price/performance.
  With several times the CI bandwidth, MEMORY CHANNEL provides a 100 MB/s interconnect with minimal latency. MEMORY CHANNEL architecture is designed for the industry-standard PCI bus.
• Requires no change to existing applications.
  MEMORY CHANNEL works seamlessly with existing cluster software, so that no change is necessary for existing applications. The new MEMORY CHANNEL drivers, PMDRIVER and MCDRIVER, integrate with the System Communications Services layer of OpenVMS Clusters in the same way that existing port drivers do. Higher layers of cluster software are unaffected.
• Offloads CI, DSSI, and the LAN in SCSI clusters.
  You cannot connect storage directly to MEMORY CHANNEL, but you can use it to make maximum use of each interconnect's strength.
  While MEMORY CHANNEL is not a replacement for CI and DSSI, when used in combination with those interconnects, it offloads their node-to-node traffic. This enables them to be dedicated to storage traffic, optimizing communications in the entire cluster.
  When used in a cluster with SCSI and LAN interconnects, MEMORY CHANNEL offloads node-to-node traffic from the LAN, enabling it to handle more TCP/IP or DECnet traffic.
• Provides fail-separately behavior.
  When a system failure occurs, MEMORY CHANNEL nodes behave like any failed node in an OpenVMS Cluster. The rest of the cluster continues to perform until the failed node can rejoin the cluster.

Copyright © Compaq Computer Corporation 1998. All rights reserved. Legal

6318PRO_002.HTML