Voice Over IP On Wireless Mesh

infractor writes "ZDNet is reporting that the Linux based LocustWorld Mesh system now has SIP routing at every node. The LocustWorld boxes have been widely used in community broadband projects where DSL is not available, so successfully that they have been seen as a threat to next generation mobile networks. With the addition of VoIP support, these mesh networks can now compete with the telcos on voice as well as data services. More details here."

Comments

Wireless VoIP isn't feasible yet...

[mindless4210]

With the addition of VoIP support, these mesh networks can now compete with the telcos on voice as well as data services.

I would have to disagree with that comment. Yes, these networks can now provide voice services, but they cannot effectively compete. In reality, wireless VoIP is still being developed and will most likely not be of acceptable quality for another year or so. Mainly, latency is the biggest issue to be conquered at this time. I think until they are able to reduce latency times significantly in these applications, it won't be widely accepted. It's just too frustrating when theres a couple seconds in between speaking and hearing a response from the other person.

Furthermore, while a mesh network can still carry a high data rate, the high number of hops to a wired connection from some locations along the network could make talking over VoIP rather unbearable. I imagine that on a larger mesh network you could experience latency upwards of 1000 ms.

Re:Wireless VoIP isn't feasible yet...

[LOL WTF OMG!!!!!!!!!]

Mainly, latency is the biggest issue to be conquered at this time.

I live in Los Angeles and communicate with an FWD SIP with which I call a conference in Japan almost daily. Latency with that is very low, and that's with a free service!

I really don't latency is the problem as much as it is making the technology easier to use for the average joe ( X-Lite is NOT easy to set up if you have router ).

Re:Wireless VoIP isn't feasible yet...

[BlueShad0w]

And on top of that, the idea of technically free VoIP calls won't go down well with the sevice providers. I can't imagine them NOT lumping enough service charges on top to make it totally unfeasible to home users, whereas the 'read more' article seems to be aimed at just those people.

Let me stick with my MSN (*awaits flames*) voice conversations with people I know

Re:Wireless VoIP isn't feasible yet...

[JoScherl]

I really don't latency is the problem as much as it is making the technology easier to use for the average joe ( X-Lite is NOT easy to set up if you have router ).

Does Joe have a router? I think not. Ok, thanks to DSL-Lines, at least here in Germany, many people get routers, but still I don't think Joe Average has one. The greater problem is that a) Joe dosn't know about it and b) he doesn't know anybody else who uses the same VoIP system. To make use from VoIP it would imho need one big company advertising these services, but I think the ISPs do not like VoIP 'cause it creates huge amount of traffic

Re:Wireless VoIP isn't feasible yet...

[mc6809e]

I live in Los Angeles and communicate with an FWD SIP with which I call a conference in Japan almost daily. Latency with that is very low, and that's with a free service!

But your situation is unlikely to be the most common.

Being on the west coast, you're probably just a few hops from a trans-Pacific link directly to Japan. You have what amounts to a nearly direct link from one place to another.

Re:Wireless VoIP isn't feasible yet...

[PretzelBat]

So long as this is internet to internet there is no service fee.

However, this sort of thing, if it becomes common, could quite possibly lead to a tragedy of the commons. If everyone actually started using all the bandwidth they had available, the networks would become jammed quickly enough.

Free VOIP is great in the short-term, but there is *not,* at this point, an unlimited amount of free bandwidth available.

Re:Wireless VoIP isn't feasible yet...

[kennybain]

In the real world, this isn't the case. You have multiple uplinks into the "wired" interent, so you are only going 3 or 4 hops into the mesh. Ping times to the internet never exceed 100ms on a properly designed mesh network. I use Packet8 over my network... http://www.fastlineinternet.com , we are the first US deployment of the LocustWorld system. So this is a voice of experience.

Re:Wireless VoIP isn't feasible yet...

[LOL WTF OMG!!!!!!!!!]

I agree with you on most of that. However, MANY people have routers. Essentially anybody who has a broadband line and more than one computer.

And routers these days are generally easy to use. CompUSSR sells one for $20 with a nice web interface and very easy instructions.

However, the fact remains that most VoIP software has horrific problems when working with a router. Whether this is problems with most routers handling UDP, or just bad programming is something that's beyond me.

Re:Wireless VoIP isn't feasible yet...

[aminorex]

Well, it's not like VOIP takes up any more bandwidth
than PSTN. If I'm calling you on a VOIP phone,
I'm *not* calling you on a PSTN phone. The difference
in backbone traffic on the fiber is negligible, but
the difference to my wallet is significant.

Re:Wireless VoIP isn't feasible yet...

[azuretek]

I work for a cable company, I'd say about 70% of the people that call have routers even if they only have one computer. Most people with routers dont know exactly what they do, in fact that's probably the reason there are so many unsecure access points.

VoIP has companies advertising and it's becoming more popular as well as usable.

Where's the beef?

[Brento]

If it supports SIP, it's not obvious from their downloads. Their ISOs haven't been updated since 2002...

Re:Where's the beef?

[AndroidCat]

There's some newer files at their high capacity mirror (Haven't checked them over yet.)

Re:Where's the beef?

[Cheeze]

i see 2/10/2004 on the latest build on their ftp site.

Quality

[lukewarmfusion]

Compete? Maybe not. Remember when NPR discussed this and one of the callers started having problems - right in the middle of his praise for VoIP?

That said, I'm anxious to find an inexpensive way to replace my $90 cell, $50 broadband cable, and $40 landline. If I can cut these bills down significantly (by using my broadband to provide my landline) I'd be happy. And I'd bet that most bill-paying consumers would be too.

Re:Quality

[YanceyAI]

They manage to allow us to consolidate those, along with the idea of the single multimedia device, then I'm happy. Very.

Re:Quality

[ytseschew]

You also need reliability. My cable modem usually goes out for a couple hours every couple months. This past week it was out 3 times while they did "service upgrades." On the other hand my phone line has been out exactly 1 time in 4 years and only because a telephone pole was knocked down right near my house. My power has gone out a number of times but that never knocked out the phone service. I can't even remember having the phone service going out at my previous location. Having my landline phone go down is so rare that it's like a surreal event. Until my internet service comes close to that kind of reliability I'll be sure to keep a landline around.

- Steve

This is precisely

[iminplaya]

the kind of "wireless internet" that I have been babbling about in other threads. This is what can liberate us from corporate control of internet access. I want to see this "wireless cloud" cover the planet. The latency issues will be worked out. In the meantime, this is great for "little" community internets where latency is not that bad. Even if they can't access the net at large, they can communicate, completely free from interference from the gov't, with each other. Maybe (hopefully) it can bring about completely anonymous, untracable communications. Just because it's not codified into law, anonymity is a right, and anything that can bring it about is a good thing.

my greatest dream

[ericbrow]

I've always thought that this should be. Wouldn't it be great if wireless networking were as easy to come by as electricty, but without the wires.

I know it's a little communistic in thinking, but I really believe that to gain true freedom of information, we need to make the information superhighway free to use.

While I know many problems would have to be worked out, like security, but it would change everything. Imagine every student being able to turn in assignments anywhere. Imagine doctors being able to monitor patients real-time, as they were being rushed to the emergency room. Yes it would put the telcos and cable companies in an uproar. But I think that would be the price of progress.

Re:my greatest dream

[cavebear42]

There should be a mod:
-1 (used phrase "information superhighway")

Re:my greatest dream

[Patrik_AKA_RedX]

And a mod -1 (or +1) "buzzword intolerant"

Queue up Lawyers and Lobbyists

[rock_climbing_guy]

Let's set up a queue for all the lawyers and lobbyists for Cingular and Nokia to try to get a bunch of stupid laws passed to tariff / cripple this technology.

So..

[bigattichouse]

question is, will it "all" eventually be one big mesh out there? I imagine the telcos will do what they can to stop it, but I could see mass mesh adoption as an incredible force to recon with.

Sign me up!

[BlueShad0w]

Great! Now find me a way to get my electricity via wireless and I can be totally independant!

Re:Sign me up!

[iminplaya]

Well, if you're near some powerlines, you can always roll up a fatty...er...I mean a really huge inductor coil. :-) Otherwise, I think that guy Tesla had some ideas on that.

Dell use VoIP for their Indian call centers

[Anonymous Coward]

5 second pause...

Hello, my name is Bob Thandushepatindiar how may I help you?

5 second pause....

My computar's borken! Help.

5 second pause...

I understand your unhappiness.

5 second pause....

I said my COMPUTAR'S BORKEN!

5 second pause...

Thank you, come again.

Latency my ass

[www.fuckingdie.com]

Latency is not a problem as far as I am concerned. I use FWD on a regular basis, and have never ran into a major issue making even ultra long distance calls. (This includes peak time calling, which has never given me trouble.)

WVoIP taking over Wireless (Mobile)

[UID1000000]

I've been thinking about this same thing just recently. From what I understand http://ask.slashdot.org/article.pl?sid=04/04/13/20 43202&mode=nested&tid=126&tid=137&tid= 215 that VoIP still isn't the most reliable thing in the world. There are a few things from a business stand point that will keep WVoIP from overtaking the mobile market, although this is perfectly tempting.

Here are a few of the reasons:

• The number of flaws and hacks that are readily available for switches, routers, and hubs and the fact that the OSes that run these appliances are too vunereable (think M$).
• The fact that the system isn't proprietary. I understand that there are ways to make a wireless network prop by MAC translation, etc.
• Handsets, currently there aren't any handsets available nor anybody in the market who wants to make them

Well, these are just a few thoughts. I know that there are many other insights as to what might make this industry grow.

Currently, to me, it seems like a lot of the open public widespread wireless networks tend to be international countries (not America) and they tend to be home grown by some geeks.

Represent a business model that would cost billions to setup and would still have to have willing hardware developers to make it happen and let me know what capital investors are interested.

Re:WVoIP taking over Wireless (Mobile)

[UID1000000]

Good point to each of those items.

What about hardware? I'm talking about load balancers, servers, and phone sets? Let's suppose you roll something like this out to a community, what would they have to have to support it? Probably a server with a 1Gb backbone right? Something to handle all of the traffic too. Then they'll need hardware for the phone replacement.
Or it could be offered as a notebook/PDA based software that you use to make your calls.

Re:WVoIP taking over Wireless (Mobile)

[grahamsz]

Well each device can be set to periodically configure itself by TFTP. That can then balance the load just by dividing the users up across your servers.

Typically you need somewhere between 32 (cell phone like) and 96 kbit/s (better than pots) quality call. However that's the peak bandwidth, at least half the call will be silence (while the other party talks) so you can survive with as little as 16kbit/s average.

A T3 should be able to support about 2800 low fi calls or almost 1000 hi fi calls. But not everyone will be calling at the same time... and also, most of the calls will be local so you wont need outbound bandwidth.

Injecting harsh realities

[PureFiction]

The code is there, the actual performance is going to be lackluster at best.

Mesh networks suffer from scaling problems due to the overhead associated with ad-hoc protocols. All that flexibility and adaptability come at a price: efficiency, latency and throughtput all decrease as the size of the mesh increases (and even more so when you have popular / power law nodes attracting routes)

Voice is notoriously sensitive to delay and to some degree packet loss. Sure, delay effects can be overblown (ATM anyone?) but you get a saturated mesh network trying to route voice and those multi-second round trip times are going to make your cable modem look like a T3.

[You get losses due to interference, transient link problems, mobile nodes, sun spots, whatever, that cause delays at the physical layer (an ethernet frame takes a while to traverse the ether) which then affects all higher layer protocols: UDP packets can't be reassembled because a fragment is lost. TCP starts backing off too agressively. Retransmission timers get triggered adding to inefficiencies, the list goes on]

Wireless and mesh networking in particular are very promising and useful technologies, but they are no where near the utopia that is often presented.

Trivial DoS attacks, scalability problems, and compounded complexity all add up to make it a very volatile environment.

Sure, this stuff will work, but only in very constrained configurations / environments.

Maybe someday further in the future these dreams can be realized when we have robust MIMO software radios and intelligent network stacks that can adapt to such harsh conditions. :-)

Re:Injecting harsh realities

[kennybain]

Actually, you are only partially right. Traditional mesh networks, such as the one that MIT has been working on, require routing tables that grow (exponentially) each time a new node is added to the mesh. However, you should read up how LW has solved those issues. From the LW website: "As each mesh node is autonomous, discovering routes on demand, there is no central control to act as a bottle neck. As the network grows the routing task for each node does not grow exponentially, as they only build routes to the resources that they need. Routes are established on demand, and un-used routes are flushed out after a short time."
(Read complete article)

Wireless vs. Fiber vs. Commercial vs. Govt

[billstewart]

Wireless clouds are cute and friendly and don't have really huge bandwidth (though I'm one of those old-timers who remembers when 56kbps *was* really huge bandwidth :-). Fiber optic pipes *do* have really huge bandwidth, and most of the locations you want to talk to are connected to them, and if you want your community wireless network to do anything useful, you're going to need to tie them together, typically using the wireless for local access and maybe cross-town access and using ISPs for long-haul backbones. You might tie into them by buying big-ISP service at some points (e.g. if you're a non-zero-cost cooperative), or community members who have friendly DSL ISPs like Sonic and Speakeasy might share their bandwidth, or you might be part of a commercial local wireless net such as Sonic.net's Sonoma County rooftop networks. The more complex and mobile your network, the harder it is to get routing right.

Interference from the government is a relatively orthogonal problem. There are several different kinds you can run into, including

• Government-run or government-mandated private companies providing local telecom or cable services that restrict what you do with your access. Wireless access that doesn't use these services usually avoids this problem, though you're still connected to a longhaul provider at some point. In some countries, it's not that easy - even though most of the world has "telecom liberalization" now, there are still usually some restrictions on competing with The Phone Company.
• Government-mandated fees (like Universal Service Gore Tax, access charges for the privilege of connecting your phone line to long distance (even if you only use it for DSL, not voice), etc. that you avoid by not using the government-supported telcos as access providers.
• Wiretapping on your data connections: Wirelesstapping is often easier to implement, but wiretappers know where the phone company buildings are, and can often bully the telco into implementing the wiretaps and passing along the cost in their rate base. Either way, if you don't use encryption, you're tappable. The US FBI is trying to expand the current regulations that let them wiretap voice calls to cover all ISPs as well as traditional phone companies, because they want the expanded power, expanded lack of accountability, and ability to force other people to pay the costs of their habit.
• Even if you're using encryption, wirelesstappers (or wiretappers, if you're going somewhere far enough away to need wire) can show which IP addresses are talking to which other IP addresses. That means that unless you're using encrypted tunnels across the wired space, and/or changing your IP address for every connection in ways that aren't logged, you're not going to get much anonymity. Building Anonymity is hard work.
• Wiretaps on voice calls - if you're using a gateway from IP SIP space into the Public Switched Telephone Network, that gateway can be wiretapped just like any other phone call. The big difference is that the gateways don't have to be located in the same jurisdiction as you are, though for cost reasons they're usually located near the recipient. That means that if you're IP-phoning one of your non-IP-telephony-equipped neighbors to plan next month's anti-war protest, your local police Red Squad can get a court order to wiretap your neighbor's phone, and John Ashcroft can get an unaccountable back door into the VOIP gateway in your area code -- but if you're IP-phoning one of your friends in another state to come speak, your local police can't tap you, because the gateway isn't in their territory, and they probably won't do the paperwork to get a tap on your friend unless you call them a lot. If you're calling somebody in Palestine to come speak, the NSA / GHCQ Echelon wiretappers might be listening, and Mossad or Shin Bet might also be listening, but that's both on the far end.
• .... many other forms of interference I haven't thought of - add your own ...

Like Air Supply, this could be big in Asia

[pangian]

The only place that my organization considers VoIP is in our offices in developing countries.

In many developing countries landlines simply aren't a viable option due to underresourced, corrupt and/or incompetent state-owned telecoms. Many of these countries have been able to develop more robust cell and broadband services, as these industries have seen less regulation and are more scaleable.

For security, convenience and efficiency reasons we like to provide staff in these offices with cell phones, however cell phones plans in may still leave much to be desired in some countries.

I think that many of our offices would be interested in VoIP cell phones if the coverage was decent (even covering major cities might be > or = to existing cell networks). Latency in phone conversations is already par for the course.

Could be an interesting microenterprise project.

Latency on Wireless; East of Use and NAT.

[billstewart]

Latency on the wireless part of a connection and latency on the wired part of a connection are much different issues, and the whole VOIP and Video-over-IP world is riddled with mythology about latency. If you've got an ISP connection with decent trunking and your uplink isn't heavily loaded, the biggest components of your latency between LA and Japan are the distance to Japan (which regular phone calls also have) and the sampling time of your codec. But if you're sharing an overloaded community wireless LAN that takes six hops through random users' PCs (anything from laptops to PDAs to Fuzzballs to 5GHz-P4s running Genome@Home) to arrive at a 384/128 ADSL connection that's competing with the Mandrake BitTorrrent and three P2P music sharing networks, you'll get a lousy connection to someone in the next town.

However, I agree with you that ease of use is the more serious problem - the prevalence of NAT routers has been breaking the Internet End-to-End model to the extent that John Walker pulled his support for Speak Freely, one of the early pioneer VOIP systems. Some closed systems like Skype do supernode things to work around it, commercial systems like Vonage and AT&T use appropriately designed equipment, and some systems limit their support to the PC-to-PSTN direction. It's an ugly mess.

Actual measurement of a LW mesh

[Baldrson]

Here are ping times from a mesh node that is 3 hops from the gateway:

PING yahoo.com (66.218.71.114): 56 data bytes
64 bytes from 66.218.71.114: icmp_seq=0 ttl=52 time=581.611 ms
64 bytes from 66.218.71.114: icmp_seq=1 ttl=51 time=231.480 ms
64 bytes from 66.218.71.114: icmp_seq=2 ttl=51 time=381.342 ms
64 bytes from 66.218.71.114: icmp_seq=3 ttl=51 time=402.864 ms
64 bytes from 66.218.71.114: icmp_seq=4 ttl=51 time=439.277 ms
64 bytes from 66.218.71.114: icmp_seq=5 ttl=52 time=412.702 ms
64 bytes from 66.218.71.114: icmp_seq=6 ttl=51 time=151.642 ms
64 bytes from 66.218.71.114: icmp_seq=7 ttl=52 time=430.497 ms
64 bytes from 66.218.71.114: icmp_seq=8 ttl=51 time=444.032 ms
64 bytes from 66.218.71.114: icmp_seq=9 ttl=52 time=280.485 ms
64 bytes from 66.218.71.114: icmp_seq=10 ttl=51 time=724.143 ms
64 bytes from 66.218.71.114: icmp_seq=11 ttl=52 time=92.999 ms
64 bytes from 66.218.71.114: icmp_seq=12 ttl=51 time=695.740 ms
64 bytes from 66.218.71.114: icmp_seq=13 ttl=51 time=419.220 ms
64 bytes from 66.218.71.114: icmp_seq=14 ttl=51 time=737.417 ms
64 bytes from 66.218.71.114: icmp_seq=15 ttl=52 time=618.897 ms
64 bytes from 66.218.71.114: icmp_seq=16 ttl=52 time=539.789 ms
--- yahoo.com ping statistics ---
17 packets transmitted, 17 packets received, 0% packet loss
round-trip min/avg/max/stddev = 92.999/446.126/737.417/183.842 ms

Here are the ping times from the gateway itself:

PING yahoo.com (66.218.71.114): 56 data bytes
64 bytes from 66.218.71.114: icmp_seq=0 ttl=52 time=64.234 ms
64 bytes from 66.218.71.114: icmp_seq=1 ttl=53 time=64.491 ms
64 bytes from 66.218.71.114: icmp_seq=2 ttl=53 time=64.086 ms
64 bytes from 66.218.71.114: icmp_seq=3 ttl=52 time=63.948 ms
64 bytes from 66.218.71.114: icmp_seq=4 ttl=52 time=63.516 ms
64 bytes from 66.218.71.114: icmp_seq=5 ttl=53 time=65.467 ms
64 bytes from 66.218.71.114: icmp_seq=6 ttl=53 time=64.871 ms
64 bytes from 66.218.71.114: icmp_seq=7 ttl=52 time=64.494 ms
64 bytes from 66.218.71.114: icmp_seq=8 ttl=52 time=64.090 ms
64 bytes from 66.218.71.114: icmp_seq=9 ttl=52 time=64.252 ms
64 bytes from 66.218.71.114: icmp_seq=10 ttl=53 time=64.044 ms
64 bytes from 66.218.71.114: icmp_seq=11 ttl=53 time=67.765 ms
64 bytes from 66.218.71.114: icmp_seq=12 ttl=53 time=64.428 ms
64 bytes from 66.218.71.114: icmp_seq=13 ttl=53 time=63.651 ms
64 bytes from 66.218.71.114: icmp_seq=14 ttl=53 time=64.078 ms
64 bytes from 66.218.71.114: icmp_seq=15 ttl=53 time=63.852 ms
--- yahoo.com ping statistics ---
16 packets transmitted, 16 packets received, 0% packet loss
round-trip min/avg/max/stddev = 63.516/64.454/67.765/0.967 ms

A caveat on these numbers. First, I haven't optimized the mesh for VoIP -- I just got my VoIP equipment in and will be getting around to that shortly. Secondly, I'm running on the mesh myself so these were output to my ssh screen simultaneously from the distant box so the traffic was doubled up.

RFC Correction

[muonzoo]

Of course, RFC2543 as referenced in the parent article is deprecated, the current base SIP RFC(s) can be considered:

1. RFC3261,
2. RFC3262,
3. RFC3263,
4. RFC3264, and;
5. RFC3265.

Re:Reality Check

[PureFiction]

Good points. What I meant by latency is that losses in the physical layer result in large latencies at the transport layer (i.e. the 802.11 MAC).

And RTP wont fragment as you mention because of MTU (unless you were doing something really odd with fragmentation at the 802.11 MAC?). I was thinking along the lines of long setup delays for the sessions due to SIP over TCP with larger payloads.

I was a bit harsh on mesh networks. The combination of AODV, DSR, and DSDV is a huge shift in the style of ad-hoc organization and cooperation that makes for a truly useful and individual/community centered approach to communication. It is going to be fun.

I just tend to get a bit annoyed with the grand visions of a nationwide mesh utopia springing up from the bowls of democracy and freedom to release us from the tyranny of Big Co Telecom and whoever else... *grin*

Hmmm, I'm going to avoid discussing security implications of the various protocols for now (that's a whole other can of worms I'm sure you are well familiar with)

Trust and security in decentralized networks makes the security problems of the enterprise look appealing in comparison :-)

Re:Reality Check

[PureFiction]

I'll shut up after this, promise. :-)

Multisecond RTT doesn't happen on anything but GPRS

I've seen it far too often on congested wifi networks. you easily get into a congested state with a crowded AP that forces lots of client waits for the DCF (i.e DIFS + padding, each in turn) and also induces lots of retransmission at the physical level due to collision with so many clients trying to talk to the same AP. Low power clients associated at the 1 or 2 Mbps rates drive this contention over the DCF even higher, severely punishing everyone associated.

The big conference venues are notoriously bad about this, as you often end up with 10-20+ people associated with a single access point. That is just too many, and the 802.11 MAC was never meant to handle that kind of load efficiently. It is a pretty good solution for the general case that simply can't cover all the edge cases (long shots, high client loads, noisy RF environments).

This type of situation results in really weird ping times, for example. I've seen fluctuations myself that go from 80ms, 120ms to 3s!, 2s!, etc. then back down to a few score milliseconds. That is the 802.11 MAC trying to cope with scenario's it was never designed to encounter.

I mentioned software radios in the first post because having access to timing and congestion control in the MAC would allow mesh boxes, clients, and AP's to make very significant performance enhancements for situations where they were needed. Why be forced to use a static, inflexible, proprietary hardware layer when you can have the open flexibility associated with software radio? (It's coming, just not soon enough :-) There are also extensions to the ad-hoc routing protocols (like passive monitor of route info between other clients in DSR) that could be supported if only the hardware was open enough to do so.

I don't want to bitch too much; we have come a long way from sub-megabit data via FHSS over 900Mhz. I just want the really good stuff to hurry up and get here already so that things like mesh networks, low latency/loss voice over IP, and highly available multipath/redundant network configurations can be enjoyed to their full potential. (software radio + multiple input / multiple output + intelligent network stacks that can handle a diverse and volatile network environment). ... and a pony!

Gratuitous links:
congestion problems at TechEd conference

congestion melt down at CeBIT

GNU Radio's software defined radio (SDR)

software defined radio on $2,000 of 'roids [it's a dev kit, but would work very well for almost any kind of project]

Re:Reality Check

[Effugas]

Wow. Very cool. I was unaware of this particular failure mode for 802.11. I would have assumed that the exponential TCP backoffs would have slowed down the incoming packet rates enough for the system to eventually slow down to some reasonable rate ... I bet, to some extent, hidden node played a _big_ factor here too.

Still, I'm amazed you saw not dropped packets, but the MAC hold onto stuff for thousands of ms. Wow.

You know, the newest Linux wireless drivers have moved _everything_ into software -- thus the ability to throw up an AP on demand. I've actually got an amusingly demented protocol hack that would address some of these shortcomings, quite transparently...any interest in collaborating? Send me an email.

--Dan