[HCoop-Discuss] Our ideal architecture?

[Adam Chlipala]

I want to summarize my take on our recent discussions about what kind of 
server set-up we should be using, in hopes that we can converge quickly 
to agreement on our targets for, say, the next six-month period.

Personally, I'm most convinced of the suitability of the following 
choice, based mostly on Adam Megacz's suggestions: buy two used IBM 
xSeries machines with more or less as much compute power/memory/etc. as 
we can fit in 1U (per machine).  Colocate them both with some provider 
that some of our members recommend.  One of the machines is intended 
primarily as a back-up in case of hardware failures in the main machine, 
but it would probably make sense to spread some services between the two 
all the time, if only so that we notice if the back-up machine breaks.  
Use Xen virtualization on both machines, so that we can create many 
virtual servers and move them among machines as needed.

This plan addresses a number of lame points in our current hosting 
situation, which seems bad enough that someone might have conjured it up 
as a straw man:

- Colocation in New York City brings an extra space/cost-of-living 
premium that doesn't deliver any extra benefit to us.  Thus, we probably 
want a colo provider outside of the Big Apple.
- There is no 24/7 remote hands in our current data center, and the rate 
for remote hands during normal business hours is exorbitant.  This isn't 
the industry norm, so it should be easy to avoid.
- We are colocating dinky machines that nonetheless take up as much 
space as beefy machines.  There is a fixed cost per U of rack space, so 
it makes sense to go for few powerful virtualized machines over many 
weak non-virtualized machines.


So, what does everyone think about this basic idea?  Issues of 
staffing/emergency response remain, but I think it's easiest to save 
those until after we agree on the hardware/virtualization situation.