Re: [Xen-users] XCP Test workstation

[Grant McWilliams <grantmasterflash@xxxxxxxxx>]

On Thu, Oct 27, 2011 at 10:07 AM, Brett Westover <bwestover@xxxxxxxxxxx> wrote:

>I got tired of ending up with hardware that costs too much to upgrade (damn you Intel) so now I buy hardware that's cheap enough to replace. The main concern however is you need to think long and hard about your CPU choice because pools like to have the same CPU in them. I like more less powerful cores over less more powerful cores so I'm building a lot of AMD hexicore systems. The new cloud will probably be AMD 8 core boards.

Well I really like that idea, and it serves two purposes for us. It would be a shift to a more efficient cost model, and its much more scalable. Now we have to fit the purchase to our requirements, and then make entirely separate plans when it comes time to upgrade. This model scales from my "cheap and easy" test requirements, all the way up to a large production cloud.

One question, when you DO need to upgrade to the next generation of processor, does that just become a separate pool? So you're kind of stuck with the CPU type you've selected, for that whole pool forever... but then come time to build a new cloud, you can make a different choice if you're requirements have changed, or the market has moved on whichever comes first. You have to build in the redundancy you require to the new cloud, and then do cold migrations of your workload to the new cloud. Does that sound right?

Pretty much although there's a certain amount of wiggle room. However unless you absolutely need all your VMs in the same pool I'd just add the new hardware and make a new pool out it. I had to do that when I went from Quad core Xeons to hexicore AMDs. The nice thing about commodity hardware is replacing equipment is fairly painless on the wallet.

 

Another question, how specific are you willing to get on hardware? I am literally looking to build a parts list in the next few days, and I would love to swap notes and get your opinion. (anyone's opinion in fact, though it seems I'm mostly talking to Grant here).

I'll give you the entire list if you'd like. My company also creates clouds for other companies of reasonable size. I have the 500 node cloud in design and a project twice that size after that. So if you folks get yourself into a bind I can bail you out on your dime. :-)

 

>"... a local drive that's currently used for nothing much..."
Last question, I'm guessing you slap a single local disk into each server just to boot the OS. What are your thoughts on using a flash disk instead? If the OS is not particularly write heavy, it would seem that this would save on cooling and power. We currently use Debian based "routers" which are really just commodity servers with 2 nics and flash disks to boot the OS image. Would XCP run well that way, or is it more dependent on its local disk?

 

The OS doesn't do a whole lot although with local SR caching this may change. My local disks just sit but to get the cost of each node down further I'll be investigating PXE booting the node. However I mentioned local SR caching which could be the fly in the ointment.

 

Thank you for all your input.

Brett Westover.

Grant McWilliams
http://grantmcwilliams.com/

Some people, when confronted with a problem, think "I know, I'll use Windows." 
Now they have two problems.

 

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users