CEO's Connection
The Elusive Database Cluster
by Rob Gardos, CEO

The significant scalability and reliability advantages of a cluster make it the preferred architecture for many IT organizations. The perils of managing a big box environment, including the nightmare forklift upgrade or idle backup equipment, can add significant overhead to any enterprise. The application and Web tiers have fully embraced this model, handling concurrent failures gracefully while scaling on demand. It works because it’s simple. The wonders of load balancers abstract the complexity of cluster management making the replacement/addition of nodes trivial. If only it was so easy for the database tier.

Databases inherently must maintain a single source of truth and be extremely fast. This makes clustering more difficult. As a result the big box is still king in the database world with all its bottleneck, cost, and redundancy implications. That’s not to say that no one is using a distributed architecture for their databases, but these are often designed to specific applications with much of the intelligence held beyond the database to handle seamless failover and growth.

While Oracle has made significant innovations in its core clustering product, Real Application Clusters (RAC), the technology has only been embraced on a limited basis. The vast majority of clusters are still two nodes and even for most larger scale Oracle RAC shops, RAC instances still represent a fraction of the broader database environment. What is holding RAC adoption back? If it truly offers all the allure of a distributed architecture, then why does IBM and Sun keep selling those large SMP boxes to support Oracle?

The culprit here is complexity. Even the most seasoned of DBAs will never get the comfort level of adding/removing capacity to manage the environment like a real cluster. In some cases, it’s the worst of all worlds since, now, there are more moving parts (more parts to fail), and the environment is still being managed like a big box. While large enterprises have the resources to get past some of these hurdles, middle-market players simply can not dedicate the energy to get this to work. Finding and retaining good DBAs is hard and once one is truly comfortable with RAC (good luck with that) it will be even more difficult to keep them.

So what’s the answer? Well let’s look at what has worked. Web/app tiers have long been clustered, and at the heart of those environments is a dedicated network appliance, a load balancer. The load balancer completely abstracts the complexity of the cluster making the process of adding/removing capacity trivial. A turnkey solution that provides the same functionality for Oracle RAC would be invaluable for driving this valuable technology into the datacenter. The key here is abstracting the complexity of the cluster so DBAs can spend their time configuring databases not systems.

Such wonders exist, and we at GridApp have helped customers migrate their untamed Oracle environment into a RAC-powered solution where capacity is managed just like a Web farm, and DBAs don’t need to worry about the next upgrade or processor failure. Imagine having these folks spend their time on looking at your data as opposed to keeping up an application.

So the elusive database cluster is in reach. Now it’s just a matter of giving up that illogical attachment to big boxes (I won’t go into the reasons behind that) and taking your database to the next level.

Robert Gardos
President and CEO

<< back