Fire, Power loss at Fisher Plaza in Seattle

Leo Bicknell bicknell at ufp.org
Fri Jul 3 20:39:57 UTC 2009


In a message written on Fri, Jul 03, 2009 at 03:22:14PM -0400, Sean Donelan wrote:
> Are you better off with a single "tier 4" data center, multiple
> "tier 1" data centers, or something in between?

It depends entirely on your dependency on connectivity.

One extreme is something like a Central Office.  Lots of cables
from end-sites terminate in the building.  Having a duplicate of
the head end termination equipment on the opposite coast is darn
near useless.  If the building goes down, the users going through
it go down.  "Tier 4" is probably a good idea.

The other extreme is a pure content play (YouTube, Google Search).
Users don't care which data center they hit (within reason), and
indeed often don't know.  You're better off having data centers
spread out all over, both so you're more likely to only loose one
at a time, but also so that the loss of one is relatively unimportant.
Once you're already in this architecture, Tier 1 is generally
cheaper.

There are two problems though.  First, most folks don't fit neatly
in one of these buckets.  They have some ties to local infrastructure,
and some items which are not tied.  Latency as a performance penality
is very subjective.  A backup 1000 miles away is fine for many
things, and very bad for some things.

Second, most folks don't have choices.  It would be nice if most
cities had three each Tier 1, 2, 3 and 4 data centers available so
there was choice and competition but that's rare.

Very few companies consider these choices rationally; often because
choices are made by different groups.  I am amazed how many times
inside of an ISP the folks deploying the DNS and mail servers are
firewalled from the folks deploying the network, to the point where
you have to get to the President to reach common management.  This
leads to them making choices in opposite directions that end up
costing extra money the company, and often resulting in a much lower
uptimes than expected.  Having the network group deploy a single point
of failure to the "Tier 4" data center the server guys required is,
well, silly.

However, more important than all of this is testing your infrastructure.
Would you feel comfortable walking into your data center and ripping
the power cable out of some bit of equipment at random _right now_?
If not, you have no faith your equipment will work in an outage.

-- 
       Leo Bicknell - bicknell at ufp.org - CCIE 3440
        PGP keys at http://www.ufp.org/~bicknell/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 825 bytes
Desc: not available
URL: <http://mailman.nanog.org/pipermail/nanog/attachments/20090703/6b6a6655/attachment.sig>


More information about the NANOG mailing list