[SAC] Virtualization hardware specifications

[Alex Mandel]

Howard Butler wrote:
> If someone has the time, interest, and is confident enough about
> hardware wankery, we need some specs and costs on a three well-appointed
> servers to replace osgeo1, osgeo2, and a few blades.  Assume we would be
> wanting to run Xen (or something like it) for a number of server
> instances on each machine (more cores and more RAM, the better).  Assume
> we're running at OSL, 2U's would probably be fine, and at least one of
> the machines should have enough hd to run our trac/drupal
> postgresql/mysql databases.
> 
> How much can we get for $20k USD?  Being the lazy and rather uneducated
> person that I am, I would head over to Dell's site and see what a
> tricked out dual-proc Nehalem with 8+ gb of RAM would be and multiply x
> 3, but I would suspect we could do better than that.  Alternatively,
> should we just outsource the hardware procurement to OSL entirely, or is
> having our own physical hardware something we really want?
> 
> osgeo1 and osgeo2 on peer1 is not our hosting future, and we are
> continually getting requests for more and more services.  Unless someone
> has a better idea, I propose we start looking seriously at OSL for our
> hosting home and decide if we want to be having our own hardware or be
> completely virtual.  Let's try to have all of the bits in place by the
> New York sprint (February or March) so we can piggyback our buildout,
> development, and deployment with that event.  Sound like a plan?
> 
> Howard
> 
I can help with the research, and highly recommend silicon mechanics as
a possible vendor. I know for my lab we just got a beefy 8 TB (RAID 6)
plus 8 core, lots of ram machine for under $5000, and most of the cost
was the hard drives.

In talking with some people I know who do this kind of thing a lot, they
recommend kvm over xen these days, though I don't know the details both
are good options. Note: Ubuntu now includes software to create clouds,
it's on my todo list to investigate how it works.

Thanks,
Alex