4 ms·

I was a sysadmin at a company that was extremely connected to some very important finance people. I can't say anything remotely specific since these guys are ri

by staunch 15y ago

I was a sysadmin at a company that was extremely connected to some very important finance people. I can't say anything remotely specific since these guys are ridiculously litigious and I personally witnessed multiple lawsuits. Honestly I didn't find it very interesting or particularly challenging from a technical perspective. I had experience running big datacenters previously and it wasn't any more or less difficult. All the normal stuff you need to do to run hundreds of servers. You have a bunch of servers running a bunch of strategies, similar to what a human could do, but faster. The strategies are tuned in real-time by the guy that created them (usually). He watches them run and either pauses, modifies, or kills them based on performance. The strategies are a mix of code and configs and run on a bunch of servers. You stick the servers as close to the exchange's (or other third-party's) servers as you can to get as much of a latency edge as possible. We had access to all kinds of advantages in this regard that normal ccompanies wouldn't get. The programming challenges were somewhat more interesting, but I wasn't super involved in that. Things like garbage collection performance tuning (so they could use higher level languages) and hard core math. But a lot of boring stuff too, like dealing with hundreds of legacy data feeds (many of which cost tens of thousands per month). The very core is just really crafty individuals sitting around all day thinking of ways of screwing^Wtaking a slice of other people's pie and turning that into a strategy which can be milked as quickly and for as long as possible.

veyron 15y ago
sysadmin stuff is pretty much standard because the challenging work is actually done by the strategists / developers. Stuff like processor affinity magic, which at one time would be covered by sysadmins, are now being covered by the people who deploy the strategies. I guess there's something to be said about the extent to which roles have melded (the line between strategist and developer and sysadmin is not clear, and in some circumstances a single person fills all roles). "We had access to all kinds of advantages in this regard that normal ccompanies wouldn't get." <-- it costs about 15K a month, soup to nuts, to get the "advantages"
- staunch 15y ago
  The star at a trading company is the traders, as it should be. Nothing has changed that or probably ever will. The programmers are just one layer closer to that than the sysadmins, which tends to make it more interesting. > it costs about 15k a month That's like saying "a piece of meat costs about $25". I don't think there was one POP (of dozens) that was the same price as another. Not to mention the equipment and connections required made the deal unique every time. There's more to it than that regardless. I'm not going to get into specifics, but don't presume to know something you would have no way of knowing.
  - veyron 15y ago
    Having bootstrapped and built a successful ultra low latency trading operation from start to finish, I can assure you that in this space the star is the technology. In other forms of HFT, such as statistical arbitrage, the traders are the stars. But my understanding is that the question specifically regarded low latency trading. To give an example, a whole cohort of traders used a simple lead-lag signal between futures and equities. Up until mid 2009, it was possible to consistently make money merely by waiting for futures to move, sending the information down from chicago (where the CME sits) to new jersey (most of the equities exchanges are housed in new jersey) and trading the relevant stocks. At the time, a popular trade regarded ES (the on-the-run emini s&p 500 futures contract) and SPY (the s&p 500 depositary receipts, designed to replicate the daily returns of the S&P 500). Here, everyone and their mothers knew that this relationship existed, and this relationship was published in various papers, both academic and professional. In this context, the "trader" who brings the idea isn't really adding any value to the process. The success of the trade was wholly determined by the performance of the platform. In a large HFT, the management cleverly structure the relationships so that the technology staff is lopped together with support and told that the traders bring value. This is intentional: there are numerous traders out there now who still believe they have value-add, when by now all of the tricks have been spilled. The quoted estimate is averaged over a year of actual expenses, whereby relationships with clearing brokers and exchanges were set up (bypassing external vendors) "We had access to all kinds of advantages in this regard that normal ccompanies wouldn't get." <-- that makes it sound like the tools to actually do ultra low latency trading are out of the reach of most people, and in reality they aren't. A competent person can build the business from 100K of personal savings :)

by staunch 15y ago

veyron 15y ago

sysadmin stuff is pretty much standard because the challenging work is actually done by the strategists / developers. Stuff like processor affinity magic, which at one time would be covered by sysadmins, are now being covered by the people who deploy the strategies. I guess there's something to be said about the extent to which roles have melded (the line between strategist and developer and sysadmin is not clear, and in some circumstances a single person fills all roles). "We had access to all kinds of advantages in this regard that normal ccompanies wouldn't get." <-- it costs about 15K a month, soup to nuts, to get the "advantages"

staunch 15y ago
The star at a trading company is the traders, as it should be. Nothing has changed that or probably ever will. The programmers are just one layer closer to that than the sysadmins, which tends to make it more interesting. > it costs about 15k a month That's like saying "a piece of meat costs about $25". I don't think there was one POP (of dozens) that was the same price as another. Not to mention the equipment and connections required made the deal unique every time. There's more to it than that regardless. I'm not going to get into specifics, but don't presume to know something you would have no way of knowing.
- veyron 15y ago
  Having bootstrapped and built a successful ultra low latency trading operation from start to finish, I can assure you that in this space the star is the technology. In other forms of HFT, such as statistical arbitrage, the traders are the stars. But my understanding is that the question specifically regarded low latency trading. To give an example, a whole cohort of traders used a simple lead-lag signal between futures and equities. Up until mid 2009, it was possible to consistently make money merely by waiting for futures to move, sending the information down from chicago (where the CME sits) to new jersey (most of the equities exchanges are housed in new jersey) and trading the relevant stocks. At the time, a popular trade regarded ES (the on-the-run emini s&p 500 futures contract) and SPY (the s&p 500 depositary receipts, designed to replicate the daily returns of the S&P 500). Here, everyone and their mothers knew that this relationship existed, and this relationship was published in various papers, both academic and professional. In this context, the "trader" who brings the idea isn't really adding any value to the process. The success of the trade was wholly determined by the performance of the platform. In a large HFT, the management cleverly structure the relationships so that the technology staff is lopped together with support and told that the traders bring value. This is intentional: there are numerous traders out there now who still believe they have value-add, when by now all of the tricks have been spilled. The quoted estimate is averaged over a year of actual expenses, whereby relationships with clearing brokers and exchanges were set up (bypassing external vendors) "We had access to all kinds of advantages in this regard that normal ccompanies wouldn't get." <-- that makes it sound like the tools to actually do ultra low latency trading are out of the reach of most people, and in reality they aren't. A competent person can build the business from 100K of personal savings :)