[HCoop-Discuss] Dedicated servers?

[Clinton Ebadi]

Adam Chlipala <adamc at hcoop.net> writes:

> Thinking about our recent discussions about VPS possibilities for HCoop, 
> it's occurred to me that something that might normally be called 
> "dedicated servers" might be more appropriate for us.  Since we can 
> easily utilize whole machines, we don't have much reason price-wise to 
> try to get our hands on slices of full machines, and it's nice to avoid 
> the overheads and uncertainties about resource sharing that come from VPSes.
>
> We had a rented "dedicated server" (fyodor) at InterServer for a few 
> years.  InterServer had some questionable business practices that make 
> me want to avoid dealing with them again, but I think it would be unwise 
> to write off the idea of dedicated servers, just because we "traded up" 
> from them in the past.  I think our experience with our Peer 1 set-up 
> has shown that we really aren't equipped to handle hardware addition and 
> maintenance at even a quarter of the speed that our members expect by 
> default.  I'm sure we could solve this by hiring staff for enough pay, 
> but I don't know that we are willing to raise dues enough to cover 
> that.  It seems like our best bet for our current membership size is to 
> be paying someone else to keep a set of machines up and running.

ISTR John Companies have a very good reputation, and what *seem* to be
tolerable pricing (if we were to go down this route):

http://www.johncompanies.com/jc_dedicated.html

They also offer dedicated colo, but $400/month + whatever the cost for a
bit more bandwidth and disk space are we could get a single machine that
significantly outclasses our entire hardware setup (mmm, 2x4
processors). As with rsync.net they offer a discount for non-profits
which I suspect they would give us since they did with rsync.net. If the
10% discount on backup space at rsync.net stacks on top of the
non-profit discount they are even more attractive.

If we got a single physical machine that powerful it would seem to make
sense to partition it into multiple Xen instances, if only because we'd
want our AFS master on another /machine/ (perhaps we could even go to a
proper setup wherein our KDC was on a separate /machine/ as well).

So then -- do we want to say that Peer1 and owning our hardware at this
point was a failed experiment (perhaps if we ever hit a few /thousand/
users...) and migrate ASAP to dedicated hardware, or do we want to
gradually migrate or even attempt to continue with our current hardware
(as the situation is only intolerable in that things are operating but
we are lacking in manpower to undertake large tasks or recover from a
disaster... demons lurking in the dark)?

If we are going to continue with Peer1 for a while, I can dedicate most
of my time starting May 27 to getting hopper online as our mail server
(May turned out to be quite a bit more hectic than I expected... but,
unless Tuesday is disastrous, the next few weeks will be open). I've
reviewed the exim config and it looks a bit saner than I thought at
first inspection. But, I'll email -sysadmin about this once I'm done
some last minute Lisping / family obligations.

If we want to scrap Peer1 and migrate to some dedicated hardware Very
Soon the question is: can we realistically do this? I don't think the
coop could handle the financial strain of $750 + $400+ per month for
more than a month or so; we *cannot* have a repetition of what happened
during the InterServer -> Peer1 migration wherein we burn money with
half setup hardware. 

We'd have to get a few more folks on board, and get absolutely
guaranteed commitments from anyone assisting with the migration to get
everything to a state where users could migrate within a few weeks, and
then get everyone migrated in at most a few weeks after that. Luckily
with AFS I think this would be a bit less of a pain that it has been in
the past; it doesn't seem like the burden on most users would be any
greater than if mire had rebooted. I at least would be willing to
dedicate significant amounts of time to this, /but/ I am not a very
skilled sysadmin anymore. 

We should see how much JohnCompanies charges for sysadmin stuff; it
might be worth it during a migration to pay them to handle some of the
hairier tasks if we can't accomplish them with volunteer time (time,
money; pick one). If their rates are reasonable it would make them even
more attractive.

I am much more open to dedicated hardware than to paid staff as it
eliminates the largest potential source of downtime for the coop while
*lowering* our monthly overhead. Not having to pay $50 for trivial
things like having a tech power cycle a machine is nice too.

So then, what to do?

-- 
Lindsay (Carlton): should i eat more post its