Re: [cobalt-users] Looking for Near 100% Up Time for a Very Vocal Hosting Customer

["Steve Werby" <steve-lists@xxxxxxxxxxxx>]

"Jerry Farquhar" <jerry@xxxxxxxxxxx> wrote:
> I hear you and even the client is pondering the cost getting up into the
> multi-thousands of $$ they even trough a very off the wall number
> of $50 K

Given the budget, if I was the decision maker I wouldn't use RaQs for the
project.  I would use hardened versions of Linux running only the necessary
services on higher end hardware (but actually probably less costly than a
current generation RaQ) and implement something like what is described at
http://www.linuxvirtualserver.org/.  Of course you may be able to implement
some other low cost solutions which the client will be comfortable with.

> Here is the even bigger kicker..  We have been hosting them free of charge
> ( something we offer to contract eminence customers only) and only had two
> outages over the last 3+ years neither of which was more than 24 hours.

But from what you've shared, I assume that there is no mail server reachable
during said outages and you don't have a plan in place to migrate the site
and email to another server in the event of an extended outage.  To me,
addressing that should be standard business practice, regardless of this
client's desires.  If that's not sufficient they need a more advanced
solution.  Since you haven't mentioned it and I brought it up already, I
assume you don't have a lower priority mail server implemented.  I would
start with that, at a different physical location of course.  Then you can
guarantee that there will always be a machine available to receive incoming
email and it will deliver the email to the primary mail server when it's
reachable.  Relatively speaking the cost is negligible.  Syncing email
between two mail servers would also be fairly easy if done every X minutes,
especially if the maildir format is used instead of mbox.  Then if there is
a long failure the users just need to switch to the secondary mail server
until the problem is resolved.  The worst case is that email received
between the last sync and the failure on the primary will be inaccessible
until the primary is available (0-X minutes of inaccessible email).  And of
course it's not transparent to the user.  The same goes with a simple
mirrored website and a low TTL in DNS (though that's nowhere near 100%
effective).  A load balancer and multiple servers is required for something
transparent.

> So
> all in all if they decide to actually change their arrangement they are
> going to go from a $0 cost to likely a $10K minimum per year cost.

There's going to be an upfront investment in hardware and the time involved
internally or by paying a consultant to setup the system, test it and train
you.  There will likely be colocation and bandwidth costs at a second
facility where you locate one or more of the servers, unless you already
have available capacity.  What you need to charge to be profitable depends
on the solution.

> Say here is one thought that comes to mind.. at one point there was a
vendor
> offering a product that connected to Raq3's physically together and
offered
> immediate fail over.  Do you recall anything about who it was and if it
> might also work with a Raq550 or XTR ?

As mentioned by Travis Ogden see http://www.antefacto.com/s1000c.htm.  I
have never seen it in action so I can't comment on how well it works.

--
Steve Werby
President, Befriend Internet Services LLC
http://www.befriend.com/