Another ZSLUG meeting report

I seem to have picked up the habit of doing at least one Lisp conference per year. This year I was hoping for an ILC (or ECLM?), but I'll have no such luck, it seems. So, the first ZSLUG-ng meeting was it and here's a summary of my trip to Zürich.

Sunday. Met the Teclo folks for dinner. The stuff they're working on is exciting in a low-level-ish kind of way — more on that later — and their team is composed by some of the best Lispers I know. (Also, as it turns out, they must have the very best intersection between Lisp and cooking skills!) Their startup vibe is intoxicating; if you're into systems programming, low-level networking stuff, et cetera, you should definitely get in touch with them.

Monday. Could see the snowy Alps thanks to the lovely clear sky. Zürich is posh, expensive, clean, and it's got great a public transportation system, mostly trams, which I much prefer to subways. Met Jorge Tavares who was in town for the meeting. Jorge was a TA of mine who pointed me at Paul Graham's ANSI Common Lisp back in my first year of college. (Note: this was before the best Lisp book had been released.) Among many things, we talked about his rather interesting research around genetic programming and evolutionary design of algorithms using Lisp.

Hans Hübner talked about his thwarted plans to replicate the Lisp Machine using FPGAs, and how he ended up implementing a Forth system instead. I was expecting some sort of tutorial or more technical details, perhaps a demo, but, alas, no such luck. Great talk, though.

Luke Gorrie then talked about Teclo. After introducing the team, he proceeded to show us how TCP badly misbehaves in today's mobile networks using insightful time-sequenced diagrams produced by their analysis tools. Some of the TCP sessions exhibited pathological behaviour, with TCP senders getting awfully confused by the odd packet lost in the ether. It was a bit reminiscent of buffer bloat to the untrained eye; perhaps because part of the problem is that radio networks go out of their way to not drop packets. There was some explanation about how Sambal sits as a proxy between radio and wired networks and massages TCP connections to make them more amenable to the lossy radio networks. They have their own TCP/IP stack written in Lisp that bypasses the OS and handles packets in under 100 ns each. Neat stuff.

The evening carried on, to a local pub, as usual. It was a good meeting, if a bit short. I think it could have included lightning talks; those have worked quite well in past conferences.

Tuesday. Had time for some light hiking around Üetliberg, then back to Lisbon.

2 komentoj:

Dave Taht said...

I was discussing the effect of wireless on TCP/ip last week on the bufferbloat mailing list.


He had visualizations? donde?

Luís said...

I don't think the visualizations were made public. Someone else has asked already; I suppose you could try too.

Anyway, you say that you "regard the wireless packet loss burst problem as completely intractable for long TCP links without using proxies." IIUC, that's the problem Teclo is tackling, with good results, it seems. :-) (I'm assuming that you're referring to HTTP proxies, etc.)