Red Hat CLUSTER SUITE FOR ENTERPRISE LINUX 5.1 Instrukcja Użytkownika Pobierz pdf (Strona 27)

Introduction to Linux Clustering

Some installations use SANs which limits the cluster size by the number of interfaces on the

SAN. However, SANs are very expensive and require special hardware.

A cheaper solution, is to build two computers with plenty of storage in them using off-the-

shelf parts and then to use DRBD to create what is effectively a HA NAS. These two storage

nodes mirror each other and can transparently tolerate either one of the two nodes failing.

These two storage nodes can then export the available storage using a network filesystem

like NFS or a block-level service like GNBD, which the rest of the nodes can use.

Depending on your applications there may be no need to have local disks in any of the

servers and they can all run directly off the network.

Here is an example for a five node cluster using DRBD for storage, providing a range of

services such as HTTP and MAIL.

Solution:

● Two nodes with RAID 5 hard drive storage in each node - “storage nodes”.

● Three nodes with no disks - “production nodes”

● Storage node are setup in Primary/Secondary mode,with LVM and ext3 ontop of the

DRBD layer with NFS exports of the data.

● Storage nodes provide user authentication via NIS/LDAP/Kerberos.

● Storage nodes provide pxelinux and DHCP for network booting.

● Quroum votes are setup in such a way that failure of both storage nodes will cause a

cluster failure resulting in all services stopping.

● Production nodes boot off the active storage node using netboot and mount the root

filesystem using NFS. All the production nodes run the same software build.

● Services are spread across the three production nodes – if any node fails, the services

are resumed on another one.

Notes:

● The above design can also be used with a small two-interface SAN. The SAN can be

connected to both storage nodes instead of using local disks and the data then

exported via NFS.

● To increase redundancy, two SANs could be used, with one connected to each storage

server and mirroring done either between the SANs themselves or using DRBD on

the storage nodes. However, standard hard drives will usually be cheaper and thus

will probably be a better solution.

Suitable Environments:

● Ideal for hosting providers – in particular, shared webhosting and email.

● Ideal for large companies to increase server availability and to centralise storage.

1 2 ... 22 23 24 25 26 27 28 29 30 31 32 33

Brak uwag

Red Hat CLUSTER SUITE FOR ENTERPRISE LINUX 5.1 Instrukcja Użytkownika Strona 27