compilerbitch: That's me, that is! (Default)
compilerbitch ([personal profile] compilerbitch) wrote2004-12-06 01:33 pm

ProgNet 2004 -- [personal profile] compilerbitch goes to the Lake District...

[identity profile] kmazzy.livejournal.com 2004-12-06 05:34 am (UTC)(link)
erm, do you think everyone should move to that thing that's number 6 to do with IP addresses?

(man i am so clever.... NOT)

[identity profile] compilerbitch.livejournal.com 2004-12-06 05:50 am (UTC)(link)
The general consensus seemed to be that IPv6 would be better than what we're using now, but it's mostly irrelevant due to the advent of network address translation (NAT) which allows many machines to share one globally visible IP address. IPv6 was created because it looked like we were going to run out of IP addresses, but that's now very unlikely indeed, so it's probably not going to change.

Actually, people reckoned that IP itself was the real problem, and that the whole thing needs to be replaced. IP works well, but is very difficult to configure, especially for home users, and is not well suited to newer kinds of network, e.g. ad-hoc wireless networks where individual devices act as repeaters, body area networks, home networks, etc. The dream seems to be that it should be possible to go to PC world/Dixons/Tescos/Mothercare/whereever, come back with a pile of shiny gadgets, turn them on and have them instantly configure themselves and work first time without any button pushing being necessary. A MP3 player should be able to find your Hi-fi via something like Bluetooth, then send your audio straight to the speakers. A PDA should be able to configure itself to use your home WiFi so you can read your email, browse the web or even watch films streamed real time.

Current hardware is potentially capable of all of this without any extra advances, but networking protocols and architectures are lagging behind. It will happen eventually, but IP has a lot of inertia so it won't be sudden.

[identity profile] kmazzy.livejournal.com 2004-12-06 06:03 am (UTC)(link)
ah yes... i had heard NAT was the preferred way forward in many people's opinions.

:)

mostly - whooooooooooooooosh over my head though.
fanf: (Default)

[personal profile] fanf 2004-12-06 06:12 am (UTC)(link)
Rendezvous is doing that kind of autoconfiguration and autodiscovery now.

A bigger problem is networking in the large - how do you manage the routing for dynamically-reattaching devices and networks?

[identity profile] compilerbitch.livejournal.com 2004-12-06 08:10 am (UTC)(link)
Yes, that was all acknowledged as being difficult but worthy of further work.

I have no idea how to fix the problem -- I was only there because the ProgNet is part funding my PhD.

Rendezvous is quite interesting -- I looked at it a few years ago when I was specialising in writing comms code for the financial markets for a living. It was hopelessly expensive (and had not-well-understood scaleability) at the time so we didn't use it in the end.

[identity profile] mobbsy.livejournal.com 2004-12-06 06:21 am (UTC)(link)
Could I have a side order of QoS and reliable datagrams with that?

[identity profile] compilerbitch.livejournal.com 2004-12-06 08:11 am (UTC)(link)
Yeah, QoS got griped about quite a bit, especially by the wireless ad hoc types.
ext_8103: (Default)

[identity profile] ewx.livejournal.com 2004-12-06 02:45 pm (UTC)(link)
NAT is bloody annoying.

[identity profile] compilerbitch.livejournal.com 2004-12-07 12:44 am (UTC)(link)
That was something that came up in one of the discussion sessions actually! There was felt to be a need for 'application level arp' that could handle route discovery in any direction through (possibly cascaded) NATs.
ext_8103: (Default)

[identity profile] ewx.livejournal.com 2004-12-07 01:52 am (UTC)(link)
I'd rather just have good old globally unique addresses.

[identity profile] panzerpenguin.livejournal.com 2004-12-06 06:21 am (UTC)(link)
I hope the food was good anyway. :)

[identity profile] compilerbitch.livejournal.com 2004-12-06 08:12 am (UTC)(link)
The workshop was held in a nice 4 star hotel on the shore of Windermere. And, yes, the food was extremely good! :-)

[identity profile] sphyg.livejournal.com 2004-12-06 05:40 pm (UTC)(link)
Fraglet = baby Fraggle?

[identity profile] compilerbitch.livejournal.com 2004-12-07 12:42 am (UTC)(link)
That was one workshop paper that was pretty cool, I thought. Fraglets are little bits of executable stack machine code, but with a twist: the execution model is based on chemical reactions. Fraglets 'react' with each other to generate other fraglets, which may possibly get sent across a network.

I have my doubts about efficiency, but the idea is actually quite neat and does lend itself to some reliability problems quite well (e.g. having more than one copy of a particular fraglet in circulation provides natural protection against one copy being damaged).

[identity profile] sphyg.livejournal.com 2004-12-07 06:34 pm (UTC)(link)
Cool, if a little over my head ;)