Joe St Sauver wrote:
>
> >Date: Fri, 16 Mar 2001 22:35:02 -0500
> >From: Joshua Wright <[log in to unmask]>
> >Subject: [P2P] How much bandwidth is reasonable?
> >
> >I have been tasked with coming up with a figure for planning purposes,
> >relating to how much bandwidth is reasonable per user in a residence
> >facility.
>
> Even modest levels can quickly aggregate to massive requirements, if you
> treat the requirements as hard guarantees and you don't oversubscribe...
>
> Assume even three thousand students...
>
> 256Kbps committment --> 768Mbps (OC12+)
> 56Kbps committment --> 168Mbps (OC3+)
>
> Or do the math backwards... if you are willing to provision a full 45Mbps
> DS3 for 3000 students, what kind of bandwidth would that work out to per
> student? ~15Kbps...
Practically every net entity is oversubscribed on the premise that not
everyone will use it to capacity at the same time; some more conservatively
than others. The provisioning is more of an "acceptable packet loss"
without dropping connections due to serious degradation. If you want
to make money at it or get formal about it, go to the frame relay
model of committed and burst rates (to your subscribers, not ISP).
As has been said before, bandwidth demand will increase to fulfill the
available supply, so at some point you must throttle.
> >Understanding that reasonable is subjective, what are other people using for
> >design guidelines? In the simplest situation, is it reasonable for a single
> >student in a Residence Facility to have 64Kbps access? Then, toss in what
> >is acceptable oversubscription and subtract the small number of students
> >that do not come to campus with a computer.
Most residents will be after Internet bandwidth. If you only have a
DS3 available, one user at 100Mbps can eat that up in theory (two DS3s
for that matter). Throw each residence in a 10Mb router port by that
philosophy.
> Of course, the reality is that:
>
> -- ANY reasonable provisioning scheme would include substantial
> overcommitment (and the larger the number of users aggregated,
> the heavier the overcommitment can practically be)
>
> -- Your load is liable to asymetric (normal web browsing load will
> resulting in inbound load levels controlling your bandwidth
> requirements; an infestation of servers running from the
> residence halls can invert that condition)
Egads, servers in dorms, externally addressable?
> -- Your load will be highly time phased; a typical pattern for
> many residence hall networks is to be hot in the evening,
> a time when bandwidth requirements from campus offices and
> labs may be low (--> you may NOT want to provision separate
> capacity for your residence halls and your main campus, given
> that the load is time phased in a way that may be fortuitous).
Agreed, and we use traffic-shaping with time-based access filters to
restrict the "filesharing hog du jour" during the day, and let them
fight it out after close of classes.
> -- A tiny fraction of users are going to consume the greatest
> fraction of the bandwidth while most users will consume
> negligible amounts; a prudent strategy, then, is to focus on
> those who consume the most, and work one's way downward from
> there (of course, this implies per port flow data for the
> network)
Quite true. We've considered asking for a "Bandwidth Hogs of the
Week" section in the student newspaper (an evil idea, admittedly),
but that could backfire.
> -- You can substantially reduce bandwidth requirements by using
> passive web caching and NAT boxes (to effectively reduce or
> eliminate unathorized servers) -- but note that a loss of
> transparency is part of the price you'll pay for that bandwidth
> control
We block only incoming connections to prevent servers.
> -- Local congestion can really "help" you -- if you are worried
> about bandwidth, shared 10Mbps hubs will result in less traffic
> than switched 100Mbps infrastructure, for example.
We started that way, but now most have 100Mbps switched (but only a
100Mbps uplink). They get campus traffic really fast, but Internet
they compete with others (unless they fall into the Napster,etc rate
limit filter).
> -- What you provision may be more determined by what your budget
> can underwrite than anything else, in the final analysis...
We can't justify a "reserved" line for residence traffic without a
specific fee for access. Currently it is free, covered under a
blanket technology fee.
Jeff Kell <[log in to unmask]>
|