You might hate Wall Street -- and who doesn't this week, but if you work in IT you owe the Street a vote of thanks because the fat cats in financial services have been a major driver of technological innovation.
Given that role, with the markets now in meltdown mode, you might think that cutting-edge technology progress would slow.
[For a deeper look at IT innovation in the financial industry, see "How IT could have prevented the financial meltdown."]
I thought that, too. But after a series of conversations with some very smart folks at Sun Microsystems and IBM, I have a different take. Bear with me as I explain why.
Latency is the enemy
From an IT perspective, the segment of the financial services market that demands the most innovation is the capital markets: the investment banks, hedge funds, and electronic markets. Where the traders work, in other words. Banking and insurance companies are major IT buyers, to be sure, but they tend to be a beat or two behind the cutting edge, says Ambreesh Khanna, the head of Sun's Global Financial Services Group.
Traders are dealing in a slice of time so thin that they are measured in milliseconds. That's infinitely faster than any human can think, of course, but much of the trading that takes place around the globe is computer- and algorithm-driven. "One millisecond can equal $100 million in trading opportunity," says Khanna. And that means the real battle is a struggle to reduce latency.
That struggle plays out across hardware, especially the CPU, the network, and the software stack. Consumers are often baffled by the push to increase the speed and capacity of silicon, but Wall Street isn't. They're the ones buying massive servers running multiple multicore chips, made by Sun, IBM, Hewlett-Packard, Dell, and others.
Before I go further, let's admit that both Khanna and IBM's Suzanne Duncan, whom I also spoke with, toot their companies' respective horns a bit when describing the cutting-edge work driven by trading.
Nonetheless, even with such self-promotion factored in, here's a real-world example of Sun's work on latency that goes beyond the obvious hardware improvements the company has developed. Solaris has always used a container model -- apps running in a virtual slice of the OS on a single server. Generally, if an application needs to send data to a second app in a separate container, the data goes all the way back to the networking card and travels through multiple layers of the software stack. The result: latency.
Now Sun's TCP Fusion short-circuits that process and moves data directly from one container to container. The result: less latency as data moves at backplane speed.