For Large Databases, Scale-Out or Scale-Up?

The eWeek Knowledge Center presents IT experts answering questions about the most pertinent enterprise technology issues of the day. This installment features Kevin Closson, Chief Software Architect, HP-PolyServe.

Q: For a user with a large database application, would you recommend going scale-out or scale-up?
A: A lot depends on whether it is a new or an existing application. If you have a really critical database running on a big SMP server and its working well, why would you want to take the risk of changing to a radically different architecture like scale-out? True, the scale-out servers will be cheaper than the big SMP server. But if you have to switch from Oracles single instance database to its RAC [real application clusters] database when you go to scale-out, you will be paying $60,000 per license instead of $40,000. Suppose you replace a 32 processor SMP machine with eight four-socket servers using dual core AMD Opterons. Oracle will charge you one license for each dual core Opteron, or 32 licenses in all. That means the list price of your database software will be $640,000 more expensive in the scale-out scenario than for SMP scale-up. You probably wouldnt do this for an existing application. But it is true that when you go from Oracle SMP to Oracle RAC you generally get the same performance from an equivalent numbers of CPUs. So if you are talking about a new application, or if you are running an end-of-life system like Solaris 9 that has to be replaced anyway, it might make sense to spend more on database licenses and save on the hardware.