Developer Only Discussion List

[Treeve Jelbert <treeve AT scarlet.be>]

On Saturday 15 Oct 2005 03:02, Jeremy Blosser (emrys) wrote:
> On Oct 09, Seth Alan Woolley [seth AT positivism.org] wrote:
> > You may have noticed recently that we've got some more advanced website
> > redirecting going on and are rebuilding much of the old website
> > infrastructure.  We haven't heard much from the colo facility so we're
> > simply moving forward as if we lost the data on the colocated server.
>
> We've got reason to believe now that we *do* have backups available, though
> we don't know how current they are.  We still don't know the real state of
> the down box and drive.
>
> > If there are any users who wish to donate server space for a more
> > decentralized mirror system that we can round-robin or who want to
> > participate in a live backup system contact us in irc or send an email
> > to Jeremy Blosser (nick: emrys).
>
> After discussing this further, we believe that our best option for the
> future is to get our primary web box on a paid line with an SLA (Service
> Level Agreement) that will eliminate the kind of unattended outage we're
> currently experiencing.  Moving away from "a donated box under someone's
> desk at work" to a paid SLA line is a next step in growing as a distro, and
> we believe we have the resources available to cover the costs.
>
> (Note that we're primarily talking about this for the web / dev support /
> application server role.  www.sourcemage, bugs.sourcemage, and probably
> eventually the SCM.  Depending on what course we pursue we will likely
> continue to use ibiblio as our primarily download.sourcemage and
> codex.sourcemage distribution site, since those things mostly need lots and
> lots of bandwidth, and ibiblio has no shortage of that.)
>
> Our identified needs for this host are:
>
> * Full control of the box.  No shared servers (and we have to be able to
>   pick to put our own OS on it, of course).
> * At least 45mbps speeds, with a transfer cap sufficient for the above
>   services (we don't have enough records to know what this would be, but
>   100GB/month should be plenty).
> * A *paid* SLA for at least the power and line, as close to 100% as
>   possible, with a recourse if there is downtime.  Donations are nice but
>   they tend to make it hard to get serious support.  We'll take discounts
>   if they're available, but nothing that compromises the level of service
>   we get to expect.
> * Physically located somewhere where we have redundant on-site coverage
>   available by experienced sysadmins in our development community.  We need
>   people available on the ground in a worst-case scenario to make sure
>   downtime is as short as possible, and we need more than one of them.
>
> While we were looking at options like dedicated business lines to
> centrally-located developer houses, I got a recommendation for a datacenter
> local to me that has *very* reasonable prices and comes highly recommended.
> They also happen to promote and support open source projects quite heavily.
> Their prices start at $59/mo for colocation of any server we want on a
> 45mbs pipe with a 100GB/mo cap and a 100% SLA.  Dedicated servers start at
> $99/mo with the same pipe and a 2 hour hardware replacement SLA.  I put out
> a feeler to them, and while they don't think they can cut us a deal on a
> dedicated server, they'll give us 15% off any colo plan.
>
> URLS:
> http://www.sprocketnetworks.com/
> http://www.sprocketnetworks.com/colocation.htm
> http://www.sprocketnetworks.com/dedicated.htm
> http://www.sprocketnetworks.com/colocation_faq.htm
> http://www.sprocketnetworks.com/dedicated_os.htm
> http://www.sprocketnetworks.com/dedicated_options.htm
>
> Look over those not just to see what their answers are, but how they
> answer.  I have it on good authority from someone who hosts there that they
> mean what they say here.  They are very flexible and reliable.  And those
> prices are too good to ignore.
>
> As for local SMGL administration, I and two other guys are within a half
> hour of that place.  The other two (Dave Josephsen (superdave) and Mark
> Bainter (shamgar)) aren't that active in SMGL development yet, but all
> three of us are Sr. Systems Engineers responsible for keeping high-volume
> corporate servers and services up 24x7x365.  None of us tolerates outages
> on our servers, and all of us know how to do what it takes to avoid
> problems or fix them when they happen.  So it's safe to say we'd be
> well-covered for redundant local administration.  :-)
>
> Given all of this, I want to propose the following options.  I'd like to
> see us select one of these fairly quickly and move to implement next week.
> I'd prefer to get some input from Kinetix on this, but I think it's more
> important we get something established ASAP, and I think it's safe to say
> after a three week outage that we can't leave things where they are right
> now even after he comes back and gets us back up.  If for some reason we
> decide this just isn't the way to go, I'd still hope we can move something
> forward next week so when he does get back we can get a solution in place
> quickly.
>
> Option A: Colocation
>
> * I would purchase and build 2 identical servers.
> * One would be deployed at sprocket, the other would stay at my house.
> * Backups would run at least daily from the live server to the one at my
>   house.
> * In case of any hardware failure, we'd have the backup available offsite
>   to replace any components or the whole thing (Dave and Mark already have
>   keys to my house to help cover the more informal servers that live here).
>   Turnaround time on a hardware failure would be the time it would take to
>   get to my house from wherever one of us was, get to the datacenter and
>   replace hardware.
> * I would also purchase a year's coverage under their 'col1' colocation
>   plan.
>
> Total cost would be around $2k for the servers (less if we tried to cut
> corners, but I'd rather get stuff we can expect to last), plus $600 for the
> first year of hosting.  We'd want to plan to replace all hardware (both
> servers) every 3 years or so, so we'd average around $1,200 per year.
>
> Option B: Dedicated Server
>
> * Hardware would be covered by them in their 2 hour SLA.
> * I'd still do backups to a machine at the house, but it wouldn't need to
>   be a hardware match, and I have stuff that could cover that on hand.
>   Recovery would be time for them to fix or replace the broken bits + time
>   to restore from backup if necessary.
> * I'd purchase a year of their 'AMD1' dedicated server option.
>
> Total cost would be $1,200 per year.  Hardware upgrades should be included
> in this.
>
> Main advantages of A are that we could probably get more hardware for the
> money ($1,000 per server is based on a 64-bit box, probably with hardware
> RAID) and we would own the hardware (which doesn't really matter to me, but
> might to some).
>
> Main advantages of B are that we need less money up front and should get
> faster response time on hardware issues since hands, parts, and knowledge
> would be on-site at all time.  We'd also be able to get it up much more
> quickly, probably before noon on Monday.
>
> I personally go back and forth on which I prefer.  Based on up-front costs
> (but not average) B is cheaper, with the only real concern being that the
> specs on that hardware might not be as nice as we might like; in particular
> we may need to ask about a bigger hard drive.  B is definitely a better
> support option IMO.  Other opinions?
>
> With either option I should be able to front the money to get it going for
> the first year, but I'd want to make as much of it back as I could, so I'd
> also like to hear from people willing to make some donations fairly
> quickly.  For option A especially that would be important.
if people in Europe are to make donations, it must be easy to transfer money 
at low cost.  For instance, I can transfer money instantly to a Citibank 
account, but almost any other type of account would entail substantial costs 
and delays.
-- 
Regards, Treeve