3.5.1 Matrix Server Administration Guide

Chapter 1: Introduction 3

same performance) than a single server. A price advantage is gained

by using commodity-level Intel-based servers instead of larger SMP

(4-way or 8-way) servers or larger proprietary filers to accommodate

scaling client connectivity demands.

• Scalable NFS performance. With multiple NFS servers serving the

same filesystems and with appropriate client balancing among the

servers, MxFS-Linux supports linearly increasing NFS performance. A

16-node NFS file-serving cluster provides nearly 16 times the

performance results over a single NFS server for the same filesystems,

up to the limit of the shared storage bandwidth.

• High availability for NFS clients. MxFS-Linux supports continuous

NFS client operation across a hardware or software failure that

inhibits NFS service from a given cluster server. Because another

cluster server inherits the same IP address used for the NFS client-

server connections, the clients will continue NFS operation to that

same IP address via a different physical server. The server inheriting

the IP address and associated NFS clients may also be an NFS server

serving a pre-existing set of NFS clients. These pre-existing clients will

not experience any interruption or delay when the new NFS clients

transition to the server. (Only the transitioning subset of NFS clients,

and not the pre-existing clients, are subject to the lock-recovery grace

period and other transitional delays.) The transitioning subset of

clients are shielded from transient errors that may be caused by the

failure that induced the transition.

• Cluster-wide NFS client lock coherency. MxFS-Linux supports

mutual exclusion with file and byte-range locks used by multiple NFS

clients connected to separate NFS servers exporting the same

filesystem(s). Note, however, that file locks via NFS are disabled by

default. Explicit administrative action is required to enable file locking

via NFS. See “Using the NLM Protocol” on page 167 for more

information.

• NFS client lock re-acquisition on server failover. When a group of

NFS clients fail over to another NFS server, any file or byte-range locks

held by those clients are released and then automatically re-acquired

after the clients have successfully transitioned to another server.