I had the chance to spend some time with the CTO of a software house that delivers into the HFT market last week. Most of the conversation was about issues that any medium CTO as a Service to large organisation faces, multiple sites across different continents. Supporting users’ desktops, the need to reduce the capital and operational expenditure and still deliver the operational excellence to the business.
When the conversation turned to the specific issues of the HFT market, my host became more animated, obviously the fun part of the job. Writing code for demanding folks who don’t accept compromises on performance keeps the design teams busy. Deploying and operating that code keeps the implementation teams even more busy and keeping the applications running across the world leaves the service team is chasing the sun. The firm has to make some tough design decisions which are expensive to work out if they make the wrong call but taking too long to take the decision is worse. A bad decision executed well is better than a good decision executed badly. A year ago they committed to Field Programmable Gate Array (FPGA) technology that would save their customers 30 microseconds, but costs a major project investment to bring to market and a year of elapsed time. You have to be brave to make the investment in the belief the market will want the solution, but wait too long and a competitor will beat you.
It isn’t easy being innovative
According to my friend, the CTO, some top HFT specific issues are:
Long deployment cycles slowing the business. It can take weeks to build the infrastructure for a project. What they want is a flexible infrastructure that can shrink and grow to match the ebb and flow of projects. Cloud like resource management from a development house is what a firm needs.
Rapid technology refresh cycles are necessary in the HFT market, but as servers get pulled out of the front line and pushed into test, then development infrastructure, the write off period is a lot less than the standard three years the accountants prefer.
Low latency networking and optimal application performance are driving them to use techniques that used to be the reserve of the High Performance Computing folks. The firm is constantly looking for the next wave of innovation.