vmMBA.com: Recent Comments

Comment on About oversubscription

web development — Thu, 05 Nov 2009 16:06:31 GMT

Quite inspiring,

This is a great article and has helped my understand why over subscription can be so important,

Keep up the good work

Comment on Some Guiding Principles for Chargeback with Server Virtualization (Part 1)

software developers — Wed, 14 Oct 2009 14:07:48 GMT

Hey, that was interesting,

this is a excellent and inspiring article, keep them coming...

Anyway, thanks for the post

Comment on Refresh Now!

web development — Tue, 13 Oct 2009 09:08:36 GMT

Interesting,

thanks for writing this article on whether or not to replace a server, but i have chosen to stick with my original opinion

Anyway, thanks for the post

Comment on The Virtualization Adoption Lifecycle

software development uk — Mon, 12 Oct 2009 09:33:49 GMT

Cool,

so what for 2009?

Thanks for writing, most people don't bother.

Comment on Some Guiding Principles for Chargeback with Server Virtualization (Part 1)

Peter Weinlein — Fri, 06 Mar 2009 20:29:41 GMT

Really great article...keep 'em coming!

Comment on Some Guiding Principles for Chargeback with Server Virtualization (Part 1)

Andrew Cooke — Tue, 18 Nov 2008 23:31:55 GMT

Great article.
I have been singing the same song when I am with customers. The challenge appears to be that we can say all these things and they listen, however we don’t appear to be offering any examples of software or any other methods of solving this problem. Sure there are always going to be 10 different ways of doing something, but some real life examples are always good.
I think vKernel recognise this, but it is too early to say if they are able to help solve the problem.
If anyone knows of any other software that fits into this space please share.

Comment on Some Guiding Principles for Chargeback with Server Virtualization (Part 1)

gerod — Wed, 22 Oct 2008 17:03:58 GMT

Chris,

Good point. I think the closest proxy for normalizing the different generations, core counts, and clock speeds of CPUs is spec.org (specifically SPEC CPU2006 rate). The challenge is that the only thing that we can see in Virtual Center or any other tool is the CPU's clock speed and core count, and there's probably no good way to extrapolate SPEC numbers within the software itself.

It would be interesting to see just how correlated the SPEC rates are with core count and clock speed. I have a feeling that they're actually getting more correlated, as CPU manufacturers have put more effort into jamming more cores onto a chip than other innovative ways of satisfying Moore's Law.

I could see how it can become more challenging now that there are fewer VMotion constraints between CPU generations because of Enhanced VMotion. Whereas before, people had separate clusters of ESX hosts of different CPU generations (and could calculate unit costs accordingly), some of those barriers have been eliminated, and it's reasonable to assume that someone could have a mixed farm of servers with different CPU generations, and could have some challenges calculating the unit cost of a CPU cycle.

Definitely an area for more consideration...

Comment on Some Guiding Principles for Chargeback with Server Virtualization (Part 1)

Chris Beauchamp — Wed, 22 Oct 2008 15:00:01 GMT

Excellent Article Gerod!

The one other component I am struggling with is, understanding the CPU efficiencies as it relates to both performance expectations and also potentially for costing factors.

Translated, newer/faster/better CPUs/machines may cost less than the previous generation but they are more efficient (note this does not directly corelate to mhz/ghz). If there was some table/chart that showed all of the CPUs in certified machines and give a relative weighting factor, we can address this in our planning and costing models.

Thoughts?

Cheers,

Chris

Comment on Some Guiding Principles for Chargeback with Server Virtualization (Part 1)

Martin — Mon, 20 Oct 2008 20:59:08 GMT

Great post! Thanx for this information. Looking forward to the next article.

Comment on About oversubscription

gerod — Wed, 08 Oct 2008 17:37:00 GMT

Upon reflection, the first part of your comment is definitely true. Deduplication isn't always a background process - e.g. for NetApp flexclones, the data is deduplicated as part of the cloning process instead of in the background. Backup deduplication typically happens before the data is sent to the backup server, so not a "background" process either.

Of course, the implementation doesn't matter so much when you're talking about the business benefits: oversubscription of resources can happen in many ways (in the case of disk, it's deduplication and thin provisioning), and oversubscription is an effective way of reducing capital cost.