On Sep 2, 7:20 pm, Jeff Liebermann <je...@comix.santa-cruz.ca.us>
wrote:
> On Sun, 2 Sep 2007 10:59:11 -0700, "Mountain Mike^^"
>
> <rukidd...@aol.com> wrote:
> >I'm just playing around with setting up a "mesh" wifi access for my entire
> >neighborhood, and would like to discuss theory if possible.
>
> Why use a mesh? The only real benifit to a mesh network is the cost
> savings on the wired backhaul to the ISP. Since you're probably only
> going to have one or two backhauls, there's not much savings. Instead,
> you get the deal with store and forward packet repeaters, that eat
> airtime, mutual interference problems, bandwidth bottlenecks, complex
> routing algorithms, and performance issues. More specifically, you
> only use mesh when you absolutely must use mesh.
>
> Light reading:
> <http://pdos.csail.mit.edu/roofnet/doku.php>
>
> Reality. Be sure to read this one carefully, especially the part
> where there's typically 50% or less probability of delivering a packet
> intact, and the reliability even at 1Mbit/sec.
> <http://pdos.csail.mit.edu/roofnet/doku.php?id=interesting>
>
> The commercial version of Roofnet:
> <http://meraki.com/products/outdoor/>
>
> >Am I right in that using a router as an AP only (or bridge) would cut the
> >bandwidth in half?
>
> Not quite. One hop will cut the *MAXIMUM* thruput in half. Let's
> pretend that you're very lucky and get a 36Mbit/sec association
> (that's a wireless connection). Maximum thruput is about half or
> 18Mbits/sec for TCP (assuming no interference, etc). However, if you
> shove that through a mesh network or store-n-forward repeater, it gets
> cut in half again to perhaps 9Mbits/sec. I say perhaps because it's
> usually more than half if there are a number of mesh nodes nearby. If
> you add a 2nd hop, it gets cut in half again or 4.5Mbits/sec *MAXIMUM*
> TCP thruput.
>
> What you need is a good overdose of reality. Setup a mesh network in
> a closed room, where everyone can hear everyone else. A dedicated
> repeater or WDS bridge will suffice. Now, try to measure performance
> from end to end using IPerf (instructions on request or Google this
> newsgroup for my posting on IPerf). If you turn off the repeaters,
> you should get something near the maximum theoretical performance.
> Turn on the repeaters, and watch the thruput drop radically along with
> a corresponding increase in retransmissions and errors. Try it.
>
> >And, for experiments sake, I'd like to get an idea of what kind of hardware
> >to buy to allow access for, say, 100 homes, in a 6 block area.
>
> Ummm... 100 customers? Every time it goes down, you want 100 phone
> calls? Have you perhaps been snorting, injesting, or smoking
> controlled substances? Never mind the technology. Do you think you
> can handle the complaints and service calls? Whatcha gonna do when
> some clueless user comes home with a virus or worm and eats all your
> bandwidth? How about the not so trivial problem that customers will
> tend to call you first, before calling their computer guru (who
> charges money), simply because you're cheaper? Do your really need
> this headache?
>
> Anyway, the loading is easy:
> 100 light email and web users
> 10 business users
> 1 file sharing user
> It doesn't matter how many access points or backhauls. One user can
> hog the whole system unless you have monitoring and traffic management
> in place.
>
> Ummmm.... How big is a "block" in feet or meters? Is that a city
> block or country block? Tall apartments of single family suburbia?
>
> >I know I'd need to convert the AP for outdoor use (I'd like to play with
> >cheaper consumer stuff), which I can do using an oudoor box and power over
> >the cat5, right?
>
> Right. It's called PoE (power over ethernet) or 802.3af. There are
> vendors that will sell you the complete package, or just the pieces.
> For example:
> <http://www.hyperlinktech.com/web/power_over_ethernet_weatherproof_enc...>
>
> >Although I'd probably design a conduit to carry the signal
> >and 12V seperately for easier maintenance. Any idea the wattage necessary?
> >I'm thinking of using PSU's from scrap computers.
>
> The delivered power is limited by the resistance of the power wiring.
> That's why PoE is normally done at 48VDC, so that the effects of the
> wire resistance is minimal. Junk power supplies are only going to
> supply 12VDC, make a bunch of noise, probably smoke the cable if you
> short it, and are not terribly reliable. Look into real PoE adapters.
>
> >Then, how about the repeaters? I need the same brand usually, correct?
>
> Repeaters are not very well defined in the IEEE 802.11 specs. The
> result is substantial incompatibility among vendors and versions.
> That's one nightmare I suggest you avoid. Repeaters are also a basic
> component of mesh networks, which I previously suggested are a bad
> idea.
>
> >I know about the LOS issues, and plan on doing all roof mounts, and USB
> >client adapters.
>
> Keep planning. You'll find that USB is limited to 16ft maximum cable.
> You can get amplifiers that will go farther, but then you have to
> supply power at the destination end. Use ethernet and PoE instead.
>
> >Also, any idea of the bandwidth necessary and how to limit it to each
> >client? (100 users, typical home use).
>
> Look at the real bandwidth managers:
> <http://www.etinc.com/index.php?page=bwmgr.htm>
> and see what you might need. Also note that it's not enough to simply
> configure a bandwidth manager and let the system free run. You gotta
> monitor the traffic to look for abuse, changes, hackers, idiots, and
> failures. You'll need to know the performance of each user and which
> one's are hogging the system. Check out various traffic monitors.
>

Just to comment on this, you really DONT need to spend your life
chasing abusers. We've been recommending per-IP controls for several
years now. But using burst controls, you can identify those that are
abusive automatically and implement effective controls that don't
need to be monitored. As tunnelling and encryption become more and
more prevalent in "abusive" protocols, you'll eventually have no
choice but to monitor overall usage, so you might as well start now
and save yourself a lot of headaches.

As an ISP you are a reseller of bandwidth. I've always has
philisophical problems with ISPs deciding which protocols are good and
which are bad. The truth is, that "good" usage is identifiable by the
fact that its not a continuous use of bandwidth. Web browsing pulls
pages and stops for a while. Even downloading movies stops after
awhile. Defining a fair slice, and then setting controls that enforce
that slice, is a generic way to manage a network without having to
care what specific users are doing with the bandwidth.

Dennis Baasch
Emerging Technologies