Red Hat CLUSTER SUITE FOR ENTERPRISE LINUX 5.1 Uživatelský manuál stáhnout pdf (Strana 31)

Introduction to Linux Clustering

8.4.2 THREE+ NODE DISTRIBUTED CLUSTER

The two-node distributed cluster detailed above will work fine with two-nodes, but what

about when three or more nodes are required?

We want data replication between the servers, but DRBD will only work with two-nodes in

primary/primary modes. This gives us two options:

1. Run DRBD between the two main servers, and the other servers can connect to one

of the two servers via a network filesystem like NFS.

This is quite simple, but has the obvious flaw of disk I/O being limited to the speed

of the WAN connection, as well as the transfers causing data cap usage on non-

flatrate connections.

Depending on your usage and requirements, this may or may not be a problem.

2. Run a distributed file system with replication support on all the nodes.

Unfortunately there are currently no distributed filesystems available which can

provide both replication and distributed data.

Therefore, the next best solution is a using a distributed filesystem, with DRBD

underneath it to provide the redundancy.

host2

host1

www.example.com

(floating A record)

smart pwr sw

sda

Internet

smart pwr sw

GSM out-of-band

management

network

DRBD across VPN

remote host

for

quorum ping

1 2 ... 26 27 28 29 30 31 32 33

Žádné komentáře

Red Hat CLUSTER SUITE FOR ENTERPRISE LINUX 5.1 Uživatelský manuál Strana 31