Re: [cobalt-users] UK Colocation

[Kris Dahl <krislists@xxxxxxxxxxxxx>]

on 7/5/00 2:59 AM, Chuck Gorish at cgorish@xxxxxxxxxxx wrote:

> these customers must also understand that constant checking on their part
> is becoming part of the problem, and that, being the nature of the
> internet, anything under 800ms is to be considered acceptable... they
> cannot control their routing to the server.. one time there may be only 5
> 'hops' while at others there may be 30.. and no one can control the quality
> or available bandwidth getting there... i may be wrong, and out of line
> here, but if i was to see my site maintaining under 500ms even during heavy
> load periods, i would not complain at all:)

Ahh. No.

Anything over 30-50ms from any major backbone is not acceptable, period.
And that is what I am talking about here... from the backbone to the
co-location facility.  Obviously a 300ms from a modem isn't too bad.  But
start counting a couple hops up--from the real network.

If you are one hop from pretty much any major backbone, on a frame relay
connection (or better yet are connected to a ISP/provider that has private
peering arangments with many backbone carriers) you should *never* get a
ping above 100ms to a decent datacenter.

If you are getting a ping above 100ms on the backbones it is because of the
following possible reasons:
a) they are buying bandwidth from the wrong people
b) it is saturated
c) they don't have enough private peering arangements... you are having to
hop several networks just to get to one that they are connected with

That is the whole point of doing private peering--to get rid of the
so-called 'nature of the internet'.

If you purchase bandwidth or co-locate with a company like InterNAP (or even
exodus) and you are 2 hops from any major backbone.  That means it is like
5ms + whatever the remote user's network is.  That effectively eliminates
'the nature of the internet'.  If the user is on a shitty network, then they
are going to get shitty access anywhere.  If the user is on a good access,
they'll get great connectivity to the site.

Incidentally the number of hops doesn't always make that much of a
difference in the network latency.  The quality of the network is important,
and things like saturation, etc., will make a much larger impact that hops.

If I ever see my site climb above 200ms for any period of time (read: 5
minutes), I will be an very unhappy camper.

-k