Moving on up

ALL THE LINUX-centered "wows" for the foreseeable future will probably be sparked by the recent Red Hat 8.0 release and stretch the never-ending debate: Is Linux ready for the desktop?

Free IT resource Hear how top CIOs turn change into a competitive advantage. Sponsored by HP Free IT resource Attend the SOA Executive Forum: Breaking SOA Bottlenecks SOAExecForum.com/may2007 Sponsored by InfoWorld

Just to be different, we would like to direct your attention to another perspective.

Aficionados of MySQL and PostgreSQL will definitely disagree -- and we will probably hear from IBM -- but Oracle is still the most popular database for business applications. Expensive? Absolutely, and the license fee is only a minor part of that cost. High transaction volumes call for robust, budget-breaking servers. If business is good, the increased number of transactions can quickly exhaust your expansion options and make those luxurious toys ready for early replacement.

The temptation to break that "forklift update" cycle by replacing those monolithic behemoths with a swarm of Intel-based commodities running Linux is strong, but may require solving three major problems: moving to an open storage architecture such as an open SAN; maintaining an efficient and compact infrastructure based on blade server clusters; and keeping administration costs within an acceptable range without skimping on performance and reliability.

The last point is probably the most daunting to tackle. Split your processing power over myriad cluster nodes, and you need the proper software to keep each node working in harmony over the SAN without the supervision of an army of administrators.

Sistina Software suggests that the latest version of its GFS (Global File System) 5.1 perfectly addresses the challenge and, judging from the specs, we tend to agree. Essentially, GFS creates a storage pool that every node, up to 256 per cluster, can share efficiently and reliably.

The latest version of GFS simplifies deployment (you don't have to compile your own Linux kernel), makes it possible to define cluster-wide user quotas, and -- returning to our earlier point -- introduces two features that seem tailored for Oracle 9i RAC (real application cluster) databases.

Normally, each computer running Oracle gets the coordinates of how to access the database from a specific directory lovingly called "home." Change any tuning parameter, or even the location of the data files, and you have to replicate that change on each cluster node that runs Oracle.

This introduces potential for errors and a significant amount of work if your cluster has many nodes. But GFS 5.1 allows Oracle's nodes to share the same "home" directory, drastically simplifying maintenance and eliminating the possibility of unsynchronized changes at the root.

If you are fighting nickels and dimes to prepare your budget for the next big iron replacement, it may be time to try a different approach -- Linux clusters. Perhaps desktops will follow.