Red Hat CLUSTER SUITE FOR ENTERPRISE LINUX 5.1 Bedienungsanleitung PDF herunterladen (Seite 29)

Introduction to Linux Clustering

solved:

● Internet connections are slow – data needs to be mirrored at both sites in a way that is

bandwidth friendly and transferred data when changes are made needs to be minimal.

● Internet connections fail/have outages fairly frequently. Any solution must be able to

handle this without split brain issues.

● DRBD is an ideal candidate for distributed clusters, but is only able to scale to two

nodes (or three using the commercial version). This causes a complication for

organisations with more than three offices that they want mirrored servers in.

● You can't float an IP addresses around a country unless you're an ISP. But even they

can't float an IP address to a server on the other side of the world.

● Fencing is more complicated – a secondary (independent) management network is

useful in order to be able to communicate to fencing devices in other locations.

Without a secondary connection, a crashed server can not be fenced if the production

network goes down, although a running server will stop the clustering if it loses

quorum so split-brain is not an issue.

How can we solve this? There are a number of solutions:

● Use a distributed, locally caching filesystem like AFS, which will cache commonly

accessed data.

● Have two main offices running DRBD and all other offices must connect using a

network file system, perhaps assisted by a proxy/caching device for improving

performance at the sites.

Devices such as WAN accelerators can be used to make optimal use of network

performance and some models have internal hard drives that can cache some data

like HTTP or SMB traffic.

● Float DNS names rather than IP addresses. When a node goes down which is

providing public services (eg: websites) have the new node providing the service

connect to the DNS server and change the A records.

The problem with this method, is the DNS changes can take some time to

synchronise across the web – this can be reduced by setting your DNS Time To Live

(TTL) to a low value – eg: 5mins – but it may not be honoured by all DNS caching

servers. (although as a general rule it is)

● In addition, it is possible to setup Round Robin DNS servers, which can allow you to

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

Keine Kommentare

Red Hat CLUSTER SUITE FOR ENTERPRISE LINUX 5.1 Bedienungsanleitung Seite 29