The snow was finally melting, day light returning, and the sun actually felt warm again. In the frigid north spring was the best season, not to be missed. One year a select few of us missed the Rites of Spring to inhabit an old, dark lab, building an amplifier for a semester-end final project. The rules were simple: use any design covered over the semester, add your own ideas, and any supplies (mostly standard NPN and PNP transistors plus discrete components) found in the lab to build the highest gain amplifier you can. The record holder, whose design was not disclosed, had a gain of around 20,000. One of the keys to success was understanding that individual components typically have a tolerance of 5%, and to really juice your amplifier you needed to sift through mountains of transistors to find ones that had just the characteristics you were looking for.

I had the opportunity to listen to Martin Thompson, the CTO of LMAX,  talk about low latency designs and matching engines. He’s promoting two memes related to performance: mechanical sympathy and the single writer principle.  The first he calls mechanical sympathy, borrowed from auto racing to describe “a driver’s ability to feel fine operations of the vehicle that include aberrations in function (such as changes in acceleration, braking rates, or vibration)” [1]. This sounds similar to what I imagine is the relationship between a jockey and his horse. Or what we were doing with transistors. Some people seem to use the term to mean going easy on the machine out of sympathy, but really it’s about understanding the properties and limits of the machine to maximally optimize its performance. Martin summed it up as “bringing the science back into Computer Science”. I look at it through the lens of engineering, but I couldn’t agree more.

References

[1] Drive Science