Add to: del.icio.us | Digg IT | Furl | Google | magnolia | StumbleIT | Wink | Yahoo! Technorati
Rate this post - TerribleTerribleBadBadDecentDecentGoodGoodAmazingAmazing
Loading ... Loading ...

Personality conflict

Prepping for an upcoming Network World feature, I had a fascinating conversation with Andrew Hillier, CTO of Richmond Hill, Ont.-based CiRBA. CiRBA writes virtualization analysis software that brings together configuration, attribute and utilization information for analysis using rule sets. Very funky stuff.

The subject of compute personalities came up in the discussion (or, more accurately, workload personalities). There are a finite number of these archetypes, according to Hillier, based on memory, read/write, I/O, CPU, etc., activity. A data warehousing system, for example, is predominantly read activity, with periodic bursts of write activity. OLTP has a more balanced read/write ratio. (Heavy duty number crunching, like calculating pi to the 10 kajillionth decimal place, doesn’t have much of either, but the CPU is pretty absorbed in the task.)

So one approach to virtualization, says Hillier, is to shack up complementary system on the same box, thus avoiding processes that spike at the same time and have a conflict over system resources. Makes sense. It’s the Odd Couple approach to virtualization; you can be reasonably sure that when Felix wants the duster, Oscar isn’t using it.

But there are other ways to make resources go further, says Burton Group’s Chris Wolfe. For example, identical VM images running similar tasks could share ROM pages, with the according performance and resource economy effects. (Call it the Conjoined Twin model.)

Both good arguments for some virtualization analysis software.

BTW, both Network World Canada and Computerworld Canada have new home pages within the IT World Canada site. Visit them.

Posted on January 16th, 2008 by Dave Webb and filed under Computer Science |

Leave a Comment

Please note: Comment moderation is enabled and may delay your comment. There is no need to resubmit your comment.