Re: [cobalt-users] 99.9% revisited...

[Kris Dahl <krislists@xxxxxxxxxxxxx>]

on 5/10/00 8:02 PM, Jeff Lasman at jblists@xxxxxxxxxxxxx wrote:

> and another:
> 
>> I have had circuits down for days.  If I had been single homed, it would
>> have spelled disaster.  One of the gripes I hear about sprint is how they
>> handle outages.........and I think thats important.  I mean, the line
>> shouldn't be going down frequently either, but if its a situation out of
>> the control of the provider, like a fiber cut by a different company, then
>> you can't fault them for it, it happened...........now its their job to
>> fix it.

This is the best example of the set.

If you are multi-homed with several major providers, there is essentially a
zero chance that each of them will be out of service.

What's the chance of Spint, UUNet, Cable & WIreless, Verio, Savvis, PSInet
all being down at the same time?  Very slim.

> So that's just a few examples of the kind of service you get when you
> spend even tens of thousands of dollars a month.
> 
> I hope you're not being a bit naieve, but I'm afraid you are.

I'm talking about the fact that if you spend on bandwidth $2500 a month
from, lets say InterNAP.  They spend probably about $500,000+ a month on
their bandwidth to something like 12 providers.  They spend at least that
much in labor for their technicians, monitoring, etc.

I don't think InterNAP has ever been dead in the water.  I can't remember
the last time I wasn't able to access Amazon.com (who has also invested
heavily in their server architecture for fail-over).  I believe Amazon
suffered a little bit when the DDOS attacks were going on earlier this year,
but by and large were still online.

About the only thing that can take the whole thing down would be a
well-placed, large explosive in downtown Seattle.

I'll say it again, if you invest in your servers, and purchase bandwidth
from a datacenter that has invested in their infastructure, there should be
no reason you can't provide 99.99 or even 99.999 uptime.  It just depends on
scale.

For us, we're still going to have a single point of failure.  We're only
planning on a single server, but this server will have hot-swap RAID, etc.,
so that it at least survives a single disk failure.  This is the level of
reliability and accessibility that we are willing to invest in.

I'm not particularly worried about losing our route.  We'll be partering
with a company that has that covered.

The weak link will certainly be the server and software we're running on my
side of the RJ45 jack.  I may actually set up one of my personal ones to
take over the primaries duties if the sh*t really hits the fans.  I'll at
least be able to drop my development server in as a production server if
things REALLY get testy.

But that's just my opinion.

-k