High-frequency trading moves compute cycles out of the cloud

I’ve written in the past about how Google locates its server farms near cheap sources of power.  In addition, Google can rely on cloud computing and move its server load so that its compute cycles are always in areas where it’s nighttime and thus cooling needs are lower and power is again cheaper. Power is a dominant cost for companies that rely on big computer installations.

So I was interested to read about the upswing in high-frequency trading (HFT) and the premium it places on low-latency in server response times. Low-latency is so important that HFT firms place their servers at or near the trading venue. So rather than follow the cheap power to The Dalles, OR, as Google did, trading firms put their server farms right in the basement of the trading behemoths on Wall Street. From RedMonk’s blog:

        “To speed up automated trading, HFT firms typically co-locate the servers that house these strategies in a data center at or near the major equity venues; firms that trade across multiple execution venues may pick a neutral data center in the middle. “Everyone is trying to take latency out of their systems — the trading venues, the ECNs, the exchanges are trying to trade faster because the broker-dealers are directing order flow based on [the venue’s] the response time,” explains [Larry Ryan, Hewlett Packard’s chief technology officer for worldwide financial services.]”

So the move to the cloud is good for commodity server cycles, but when every millisecond counts, in the words of Chinese general & military strategist Sun-tzu, “Keep your friends close – and your [servers] closer.”