There is an interesting story in a recent newsletter about trying to find the location of an exchange's matching engine to aid in picking a colocation facility. I'm including the story below. I'd be interested in comments; my thoughts:1) I've never heard of an exchange keeping the matching engine's location secret.2) Worrying about single microseconds seems extreme, have the black box wars really gotten to that level?----original story----The undercover performance team had been working for months now, and they were not even half way through the project. First, of course, there were the logistics to arrange. Temporary rentals of servers in a number of different locations, arranging the movement and installation of the servers, ensuring they were calibrated for finest resolutions, connecting to the downstream feeds, many other details ... it was an unusual logistic and technical challenge for a performance team. The first results were beginning to come in now, and already results were sufficient to identify where the next wave of servers would need to be targeted.What they were doing was perfectly legal. Indeed, it would have been quite a lot cheaper and faster to have obtained this information illegally. But management were quite clear that anything illegal was completely, 100%, out - no illicit corner cutting was to be allowed in any way. The project was reviewed in detail on a weekly basis by a legal support team to ensure that even fuzzy legal boundaries were not approached.So, what was the ultimate goal of this secret undercover performance project? A microsecond. All this effort over just one microsecond! Well, not actually a single microsecond, it was a microsecond per message - indeed, potentially several microseconds per message. Is a microsecond delay in receiving a message important to to you? It was to this team. You may have guessed by now, that the team worked for a hedge fund; that the connection they were measuring was to the trading exchange that they ran black-box trading applications against; and that a one microsecond faster connection to the exchange would be more than enough to cover the large costs of this undercover performance project.The actual exchange location is a closely guarded secret. By testing the connection latency from various different locations, eventually the team would narrow down to the building where the exchange was located - and then install servers running their black-box trading applications there. Every 300 meters further away from the exchange was one microsecond longer in latency to and from the exchange - and every microsecond difference in latency was worth millions of dollars to this hedge fund.

Makes sense, although they have the physics slightly wrong. Light only travels 300 meters/microsecond in a vacuum. In an optical fiber, the speed is down around 200 meters/microsecond.I'd also bet that latencies in the algorithms at the exchange, counterparties, and the HF make a bigger difference than knocking another microsecond off the one-way signal latency.That said, I'm sure this HF slightly improves the probability of getting good fills and certainly improves the probability of getting investor money.

The story is apocryphal but a bit of a cartoon. I am not a line worker for the phone company. But I can take a guess that the fastest location is not physically close to the building where the exchange server is (unless they are trading by radio), but rather the closest hop to whatever router drops them onto the exchange's switch. Rather than being a single location, this is probably more like a radius, where hedge funds all over the same city are at equal "distance" just by luck.Moreover, I would bet a typical monthly fiber infrastructure project, or a simple flip of a switch at phone-company headquarters to adjust loads, could shuffle which hedge funds are in the first and second rows at random, any minute of any day.Getting in the right city helps. But beyond that, I would imagine the best strategy is for them to simply keep all their test servers in place, and then use whichever ones happen to be fastest from minute to minute.

This story sounds a bit fishy. Any upgrades to infrastructure, changes in switches, etc.... could affect the latency issues here. Thereby, changing any throughput speeds at the "fastest location". However, I could see the case where they had servers running in different locations and constantly monitored the servers for fastest throughput. Then, sent through the fastest location.Where was this story posted?