HP TruCluster Server V5.1B-1 Specification Download Page 2

Page: 2 / 19

Cluster members in a given cluster must all use Memory Channel or must all use LAN. These interconnects cannot be mixed for cluster communication in
the same cluster. Using multiple shared buses and redundant Memory Channel or LAN interconnect hardware promotes no-single-point-of-failure (NSPOF)
characteristics for mission-critical applications.

A TruCluster Server cluster acts as a single virtual system, even though it is made up of multiple systems. Cluster members can share resources, data, and
clusterwide file systems under a single security and management domain, yet they can boot or shut down independently without disrupting the cluster.

Cluster File System

The Cluster File System (CFS) makes all files, including the root (/), /usr, and /var file systems, visible to and accessible by
all cluster members. It does not matter whether a file is stored on a device connected to all cluster members or on one that
is private to a single member. Each file system is served by a single-cluster member; other members access that file system as
CFS clients with significant optimizations for shared access. CFS preserves full X/Open and POSIX semantics for file system
access and maintains cache coherency across cluster members. For instance, an application can use standard UNIX file
locks to synchronize access to shared files. The member that will serve a given file system can be specified at file system
mount time.

For higher performance, applications can use direct I/O through the file system to bypass the buffer cache. CFS also
provides a load balancing daemon to monitor and analyze file system usage. The daemon can be configured to
automatically relocate file systems based on CFS memory usage, or during changes in the cluster, such as when members
join or leave, or when storage connectivity changes.

CFS supports the Advanced File System (AdvFS) for both read and write access and supports AdvFS with BSD-type user and
group quotas. NFS client and NFS server are supported for both read and write access, and NFS services are accessible to
clients through cluster aliases in addition to the default cluster alias. The UNIX File System (UFS) is supported for read and
write access from the CFS server or for read only access on client members. The Memory File System (MFS) is supported for
both read only or read and write access by the member on which the file system is mounted. Remote access and failover are
not supported. The CD-ROM File System (CDFS) and Digital Video Disc File System (DVDFS) are supported for read access
only.

The Cluster File System (CFS) load balancing daemon (cfsd) can monitor and analyze file system usage, make
recommendations, and automatically relocate file systems. The daemon can be configured to automatically relocate file
systems based on CFS memory usage, or during cluster transitions (when members join or leave the cluster), or when storage
connectivity changes. In addition, the member that will serve a given file system can be specified when that file system is
mounted.

Device Request Dispatcher

Device Request Dispatcher The device request dispatcher supports clusterwide access to character and block disk devices, and to tape and tape

changer devices. All local and remote cluster disk and tape I/O passes through the device request dispatcher. A member
does not need a direct connection to a disk, or tape, or tape changer device to access data on that device. This permits
great flexibility in selecting a hardware configuration that is both economical and useful.

Cluster Alias

A cluster alias is an IP address that makes the cluster look like a single system to clients and other hosts on the network.
Cluster aliases free clients from having to connect to specific members for services. If the member providing the service goes
down, a client reconnects to another member elected by the cluster alias to provide the service. With applications that run
concurrently on multiple members, scaling is achieved by permitting multiple clients to connect to instances of the service
on multiple cluster members, each using a cluster alias to address the service.

The cluster alias subsystem provides an optional virtual MAC (vMAC) address that can be associated with each cluster alias
IP address. When configured, the same MAC address is used in all Address Resolution Protocol (ARP) responses for the
cluster alias address, independent of which cluster node is responding to cluster alias ARP requests. This permits faster
failover when a new node assumes responsibility for responding to cluster alias ARP requests.

QuickSpecs

HP TruCluster Server V5.1B-1

Features - TruCluster Server V5.1B-1

DA - 11444 Worldwide — Version 4 — December 8, 2003

Page 2