Setting Up a Test/Learning Environment

Before you decide to implement clustering in your production environment, you should set up a test environment so that you can educate yourself on the practical aspect of clustering.  I will try to outline an economical test environment of 2 servers using a shared SCSI link between them to allow you to test almost any scenario for application failover.

You will need 2 fairly powerful systems to do your testing with since you don't want to waste hours waiting for the computers.  Luckily, today you can build a very powerful workstation with internal IDE disks for under $700.00 US each - or buy them from a major vendor for under $1000.00 each.  My recommendation is as follows;

The servers should have all the hardware installed and working.  The SCSI adapters on each server should be set to 2 different ID's.  This is usually done in the SCSI card BIOS - I suggest ID6 and ID7. 

Then you install whichever OS you are testing with,  ensuring that all the hardware is working before you plug in the external hard disks.  When you connect your external hard disks, you connect one cable to each computers SCSI adapter.

When you install your clustering software, each clustering software vendor treats the shared disks differently. Microsoft Cluster will use one disk as a quorum disk where all cluster information is stored. Veritas Cluster will use one of the disks for a heartbeat disk. If you use Microsoft or Veritas Cluster you should make sure that you have one extra smaller disk (or separate physical partition if you are using a Fibre Channel Array) to use for these. If you are using any Legato Product (Cluster, Octopus, Standby Server, Availability Manager) or Double-Take, you will not need this extra disk. Also, If you use Double-Take, Octopus or Standby Server, you will not even need a shared SCSI disk as these products support block level mirroring of files and partitions between cluster member servers. In all cases, your Cluster Software Documentation will step you through setting up and assigning your drives.

There are two network cards in each server. One should be connected to the main LAN and the other should be connected to the other server that is a member of the cluster - via a second hub or a crossover cable. This second network should be on a separate subnet from the main LAN and routing should not be turned on so there is no communication between the two LANs. Your WINS,DNS and other services that will be running on these servers should be bound to the main LAN (if bindings are applicable).

There are two ways that Clustering Software handles your services. MS Cluster, Veritas Cluster, and Legato Availability Manager (formerly Legato Cluster and before that Fulltime Cluster) all use virtual groups of resources such as IP addresses, network names, and services which they move between cluster members. Octopus, Standby Server and Double-Take use a primary/slave relationship between servers. When one of the servers fails, the other will take over the name, IP address, name and services that were running on the other cluster member server. Both servers can be a primary and a slave at the same time for two-way failover.

That is it in a nutshell.  This will give you a basic clustering system that you can use with virtually any clustering software to educate yourself.

If you have any comments, concerns, or suggestions, please send me an e-mail and I will respond.