Sunday, September 9, 2012

5 milliseconds!

Would you take this one on as a project?
  • There is an existing legacy capability
  • It takes about 46 milliseconds to execute a transaction with the legacy
  • Your project objective is to reduce this by 5 milliseconds.
Doesn't sound too bad, improving things by 10%.

However, in a posting by Azimuth, we learn this project is about straightening the route of Atlantic submarine cable from Halifax, NS to London so that it's 310 miles shorter (5 ms in electronic terms)
It's all about flash trading, to be made flashier still:
In fact, the battle for speed [on Wall Street and in The City] is so intense that trading has run up against the speed of light.
For example, by 2013 there will be a new transatlantic cable at the bottom of the ocean, the first in a decade. Why? Just to cut the communication time between US and UK traders by 5 milliseconds. The new fiber optic line will be straighter than existing ones:
“As a rule of thumb, each 62 miles that the light has to travel takes about 1 millisecond,” Thorvardarson says. “So by straightening the route between Halifax and London, we have actually shortened the cable by 310 miles, or 5 milliseconds.”

That's interesting, but not too threatening. On the other hand, Azimuth goes on to describe stuff that has real risk, especially given the performance of the software project recently put on line by Knight Traders:
But that’s not all. When you get into an arms race of trying to write algorithms whose behavior other algorithms can’t predict, the math involved gets very tricky. In May 2010, Christian Marks claimed that financiers were hiring experts on large ordinals—crudely speaking, big infinities!—to design algorithms that were hard to outwit.

Yikes! I hope they do a bit of structured requirements analysis, the old fashioned kind (sorry, Mr DeMarco!), when we wanted to know if it would really work. And, more better, how about a FMEA (Failure Mode Effects Analysis)? This is kind of thing NASA did when there was more than just money on the line.