Datasheet

“main” (Installation and Administration) — 2004/6/25 — 13:29 — page 395 — #421

High Availability under Linux

be mirrored one hundred percent. This normally requires clients that

have connections with two or more servers at the same time and that

communicate with all of them. A normal web browser cannot do this.

Load Balancing The distribution of load within a cluster of computers.

Load balancing is used in an LVS scenario (Linux virtual server), for ex-

ample (see Section 19.5.2 on page 399).

STONITH Shot the other node in the head: Special hardware and software

that ensures that a faulty node does not write-access distributed me-

dia within a cluster, threatening data consistency in the entire cluster.

This involves simply disconnecting the system from the main power

supply.

19.2 A Sample Minimum Scenario

The procedures within a two-node cluster when one node fails and the var-

ious types of standby systems that can take over as necessary are outlined

below (see Figure 19.1 on the next page).

The two servers (primary and backup) are both connected to a SAN (stor-

age area network). Depending on the mode, this is only accessed by the

active node. The servers communicate with each other in such a way that

they regularly emit a “sign of life” (heartbeat). The communication chan-

nels (or heartbeat links) are also laid out in a redundant way, so indepen-

dent channels can be used by means of a variety of network cards and

cable channels. If one of the links fails, its backups continue to report cor-

rectly that the relevant server is still “alive”. If there is no sign of life from

the main system, the standby system is activated, so it takes over the ser-

vices of the failed partner and removes it from the network completely

(STONITH).

19.3 Components of a High Availability

Solution

A high availability solution consists of several different components:

General Infrastructure When designing a high availability solution, it

should generallxy be remembered that even the installation of all key

servers at a single location can be a potential SPOF if this location is

395SUSE LINUX Enterprise Server