Blueeyed Devil

Monday afternoon:

Born to be a god among salesmen
Working the skinny tie
Slugging down fruit juice
Extra tall, extra wide

"Blueeyed Devil", Soul Coughing

Lunch time I talked with a gov't contractor who was in on the Hadoop tutorial. She talked about using a filesystem that was forty years old -- yes, that's a four zero -- which had lots of "warm" data (her term; I assume between hot caching and archiving to tape) cached to tape, but done very badly. The directory structure needs to be preserved, perhaps not at all costs but nearly; there are instances of old (maybe not 40 years old but close) documentation that refers to old paths that must not be broken. Interesting problem.

Also heard about this problem, which just gobsmacked me with its fullbore crazy.

Also, from other quarters, heard about a lab that lost its funding, which leaves it in a difficult position as it has a crapload of old G4s or G5s, watercooled, about half of which they discovered are leaking...

(Trying not to turn into Perez Hilton here. Not sure how well I'm doing.)

In the afternoon I took the Packaging for Sysadmins tutorial, which would have been much better (IMHO) handled as a hands-on workshop. I came back for the second half, but honestly it was a close thing...and yet when someone asked him, the instructor dropped gems of info about Func and Cobbler, which I'm going to be looking into as soon as I can.

During the break I talked with Derek, who's a sysadmin at a NYC trading firm. This was an absolutely fascinating talk, and only partly because I wasn't really aware of the whole high-frequency/low-latency trading...um, culture? algorithm? So:

• He has small data centres -- like, racks -- scattered across NYC in order to be rilly rilly close to the exchanges. Also works well for redundancy. The colos that are close to the exchanges are filled with fellow trading firms.

• The idea is that if you get your data from the exchange soon, analyze it soon, then get an order back to the exchange soon, you can make a lot of money. As a result, a 2.5 ms difference, like in swimming or 100-metre dashes, is absolutely huge.

• Improvements in speed are looked for all over the place: RT Linux, not running NTP on machines (partly because of the overhead it introduces, and partly because it doesn't have high enough resolution; better to take timing info from ethernet frames, which'll get you down to 7.03 nanosecond acccuracy), RT Java (which I didn't know existed), and even running apps on switches that run Linux (which, yes, may be slower than big servers, but are that much closer to the exchange and so it makes up for it0

• So his server rooms are small-ish and many (which if I was a better man I could turn into a full-on Dr. Seuss book), but get this: a trader's desk will have four workstations at it, each with four big-ass monitors sitting on top of their desk so that they can monitor the stocks they're trading. His power and cooling issues are at the desktop as well as in the server room. Madness.

After that back to my room, where my roommate (who's British) and I wondered at the madness of looking to the UK Conservatives for relief from a right-wing Labour agenda. Madness upon madness.