[ previous ] [ next ] [ threads ]
 From:  Andreas Bahr <eyebear at prima dot de>
 To:  m0n0wall at lists dot m0n0 dot ch
 Cc:  Chris Bagnall <m0n0wall at minotaur dot cc>
 Subject:  Re: [m0n0wall] Estimating bandwidth for net access
 Date:  Sun, 20 Aug 2006 12:33:02 +0200
Chris Bagnall schrieb:
> Greetings list,
> Slightly off-topic, but since there are plenty of folks here who do fairly
> large network installs, I hope it's relevant.
> How do folks calculate the size of a pipe required for net access? Is there
> a particular formula you use?
> When I've been asked to work out details for such things in the past,
> they've been in areas where my answer has always been "the fastest
> available". However, I've just been asked to work out a proposal for a
> 360-seat site in a major city, so options for net access and sizes of pipes
> are far wider.
> Any thoughts?
> Regards,
> Chris

This is black magic.
For a larger network we calculated that a WinNT-Workstation 
needs a 64kbit bandwith in both directions to access the 
For Win2K and XP, we  were advised to calculate with an 
256kbit assumption. You can find these numbers also in the 
Tech handbook...
Thats it for the workstations. The servers can't be 
calculated in this manner, that depends on the trafic the 
clients will create. Things to keep in mind: SMB doesn't 
saturate the bandwith via enlarging the TCP window size, its 
doing a time division thing.
If you need AD replication between several servers on 
different locations: give them time to do their job. They 
send smtp-messages (Exchange also) and hustling with the 
replication will damage your replica.

For testing we limited the bandwith between 2 GC servers to 
10 mbit, and  the replication went on as smoothly as with 
the normal 1gbit connection. So, my assumption on this is, 
that a *MINUMUM* of 10 mbit is fairly enough for connections 
between  servers.
But keep in mind: This is black magic, and if you want that 
your assumptions to work in the expected manner, you should 
sacrifice a black goat/sheep/cat at fullmoon.