Linux clusters have been a staple in science labs for more than a decade, but enterprise IT shops are just cutting their teeth on them. When a cluster problem is too tough to swallow, IT pros turn to our resident Linux cluster and administration expert, Sam Greenblatt.
Greenblatt is senior vice president and chief architect for Computer Associates' Linux Technology Group. Here's a sampling of his advice to IT pros who needed help with clustering conundrums.
How do I back up Linux clustering file systems?
Sam Greenblatt: In most cases, the best choice is to backup the cluster from a remote machine with client agents installed on each machine in the cluster. This will take the resource load off of the cluster itself. The alternative to doing remote backups would be to install backup software on the cluster. The two types of installs would be to either install backup software to each node in the cluster or to install at each node to the same place on a shared volume, and configure for failover. If you install backup software to each individual node in the cluster, then it will be configured the same way non-clustered machines are configured. So, whether your have two machines and two tape drives, or you have a fibre library and plan on using a SAN/NAS and want to take full advantage of your resources, then you would configure backup software as if the two machines were standalones. This reduces the load on the network to the remote agents. The key is trading speed for simplicity.
I have a network of about 40 Linux workstations which access the Internet via two T1 connections. Is there some way to configure them so if one gateway goes down, they will automatically switch to the other gateway?
Greenblatt: Configure two clusters into one virtual box, and share one IP address. You'll need to use dynamic routing on the client. Many networking products have failover capabilities built into them. Using software is the more effective way to solve the problem, because you don't have to worry about reconfiguring clients correctly to deal with the redundancy. Point them all at your one gateway IP, and let the router software do the work.
We're looking at putting our storage on a Linux cluster, but I've heard from some vendors that managing a Linux cluster can be challenging. (Of course, the vendors want to sell us their management products.) What are the management challenges involved in running a Linux cluster in a heterogeneous enterprise environment?
Greenblatt: Clusters require hardware-independent, well-supported solutions for Linux cluster management. The software needs to be able to integrate with and complement existing Linux solutions to provide a comprehensive solution with workload management, system monitoring and administration, and provisioning (installation and configuration) capabilities.
You will need workload management software, resource management software, and server provisioning capabilities, as well as system monitoring and administration, load balancing, high availability, job scheduling and remote Web access to manage distributed workload processing across the whole cluster.
What are the steps performed by POSt in Linux?
Greenblatt: The post-installer is a utility in Linux that will set up your database installation, according to the Red Hat database installation guide. If you chose the Server Class option or if you installed the Server package through the Customized Installation option, the post-installer runs after the installation of the packages is completed. The post-installer will, if you enable it, initialize a database cluster. After the post-installer initializes the cluster, you can perform any normal installation task.
For more information:
Check out Sam Greenblatt's Ask the Expert section for answers to more tricky IT questions.
Got a clustering or administration problem? Ask Sam Greenblatt!
This was first published in September 2004