generators, etc....

Sun Oct 13 02:17:22 UTC 1996

On Sat, 12 Oct 1996, Zachary DeAquila wrote:

>   Sure, Mike, but how do you protect against an airplane falling out of the 
> sky?  or having the building that houses your generators flattened by
> a runaway semi?  Or the ever present possibility that the building next
> door will have a gas leak and explode?  And what about that house-sized
> meteor that could come hurtling down? 

I suppose you think this is funny. But the people who run datacenters for
large corporations (like insurance companies) and important government
operations (like the taxman) do take these things into consideration.
That's why you find redundant locations (like muliple exchange points) and
data centers that are located two stories underground. About the only
scenario you mentioned that the underground data center is vulnerable to
is the metoer and a baseball sized one would likely suffice to destroy a
whole town. That's why it is wise to not have everything at one physical
location. Redundancy, redundancy, redundancy.

>   Give me a break.  Hindsight is 20/20, it's easy to see how things could
> have been avoided, 

That's right, so use hindsight to make better plans for the future.

> but excessive paranoia can and does get in the way of
> getting real work done.

Not at all. Paranoia is for the people who make site plans and who
reccommend site planning issues to management. It doesn't need to consume
your attention all day long. Just be ready when the boss comes in for a
tour, point at a box in the corner and say, "See that box there,
if it breaks then the entire Northeast would be off the air for 24 hrs".

>  Any engineer worth his salt will tell you that
> 100% reliability is unattainable - IMHO, these days with the technology
> we work with daily as young as it is, I'm impressed with 90% uptime...

I'm not. Five nines quality *IS* attainable and the telcos generally
manage this. Maybe individual components or subsytems will have as low as
90% uptime, but the entire mesh can be engineered for 99.999% uptime
even with unreliable components like that. Five nines is equivalent to 
8.76 hours downtime per year and that includes scheduled events.

> For all the effort you put into saying how you could have done better,

I don't recall saying that I could have done better. I do recall saying
that we (the industry as a whole) can do better in the future. Rather than
throw up our hands when these events occur and say it's just bad luck, we
can use them to learn where our blind spots are and fix the problems.

> I hear the goverment has an installation that might meet your standards
> somewhere under Cheyenne Mountain....

I think that installation has much better than five nines uptime. What's
wrong with learning from their example? If organizations like the Freemen
and the OK City bombers weren't such frigging idiots they could probably
destroy Western civilization as we know it by knocking out most of the
USA's key power and communications infrastructure. Modern technological
civilization is built on a house of cards and it's about time we started
hardening the foundations before it collapses.

Michael Dillon                   -               ISP & Internet Consulting
Memra Software Inc.              -                  Fax: +1-604-546-3049
http://www.memra.com             -               E-mail: michael at memra.com