Kevin Closson's Blog: Platforms, Databases and Storage

I find myself in the most interesting conversations. You know us propeller heads; we’re always talking about speeds and feeds. I find myself surprised quite often how many Oracle-minded folks give any credence to CPU frequency. I go beyond surprised and fall off my chair any time the same folks give any credence to Linux bogomips values.

They’re called bogomips for a reason.

Here’s a secret formula when running x86 instructions:

Bogomips = cpu_MHz * 2.0012069226

Yes, the formula for PowerPC and Itanium is likely different, but don’t point PPC/Itanium Linux users out in a crowded theater because it might have the same effect as yelling, “fire.” Ok, all kidding aside, I don’t have any PowerPC stuff. I wish I did, but I don’t.

That could be the end of this blog post. I’ve said enough. Bogomips is 2 times the clock rate. What a boring blog entry.

Oracle is a Load and Store Engine

…and nothing else matters. Quite some time back-when I had an occasional 5 minutes to make a blog entry-I had a series of blog posts about Oracle on NUMA. In that series I offered up a microbenchmark called the Silly Little Benchmark. That post contains a URL to the kit where it sits on an OakTable Network server. I’m using a slightly modified version at this point now. I need to update the version that is there, but since it has only been downloaded 440 times it might not be of interest. Nonetheless, I’m tying it in to this thread.

The point is that a workload like SLB will show you a great deal more about what Oracle will behave like on a system that does anything related to CPU frequency, and that includes bogomips for certain.

Just How Bogo is Bogo?

This is the first in a short series aimed at pointing out just how bogo bogomips are. In the meantime, I recommend taking a peek at the output of dmidecode(8) on your system and make sure you have symmetrical memory populating the DIMM slots on your system. An oddball configuration can cause performance degradation.