3 ms·

what does this do? (non-finance guy here)

by DevX101 12y ago

jackvalentine 12y ago
The command introduces a delay of between 100 and 10ms to every packet. Screwing up trades that rely on being the fastest. http://www.linuxfoundation.org/collaborate/workgroups/networking/netem http://www.linuxfoundation.org/collaborate/workgroups/networ...
noonespecial 12y ago
Its a bit of a joke. Its a network thing, not a finance thing. What it does is adds a delay and more importantly, a bit more random delay to the time it takes the order to reach the exchanges server. Once you add in the non-determinism, HFT basically falls apart because you can't take your truckload of cash and buy yourself a place in a datacenter that's 2ms closer to the exchange and front-run everyone. It would be a wonderful thing to see all of those millions these guys have "invested" shaving a millisecond or two off their transaction times laid waste by a single command.
- dzderic 12y ago
  This "solution" will only make it harder for regular folks to execute orders, since HFTs will beat the randomness by shooting multiple orders through multiple order gateways.
  - noonespecial 12y ago
    Guess we'll need a hierarchical token bucket with stochastic fairness queueing as well. We don't just need it to be random. We need there to be no way of ever quite knowing if any given order will beat another order to the exchange (within a given time period, of course). They won't know if they can beat joe ordinary, and they definitely won't know if they can beat the other HFT's. That might be enough to put a lid on it. Edit: For those playing along, here's the metaphor. Joe goes to market to buy sheep. Bill knows Joe is going so he sends a fast runner ahead of him to buy the cheapest sheep in town first so he can mark them up and sell them to Joe when he arrives. We try making everyone wait at the town gate for a random amount of time to give Joe a chance to arrive and get through. So Bill (being very rich) just sends 10 guys so one is very likely to be let in before Joe anyway. Next we introduce the stochastic filter. We make everyone line up and then shuffle the order every once in a while, but Bill still has more guys so he might still get one in first more often than not. Finally, we add the token bucket. For every one guy that we know employed by Bill admitted, we make the next one wait twice as long to get in, so if Joe and 10 Bills show up, Joe and the first Bill are essentially on even footing again because the 2nd through 10th Bill would have to wait too long to matter.
    - gd1 12y ago
      "For those playing along, here's the metaphor. Joe goes to market to buy sheep. Bill knows Joe is going so he sends a fast runner ahead of him to buy the... " Seriously, how many times have we discussed this issue on this site and we still get this bullshit. Bill doesn't know Joe is going. He doesn't. Get it through your thick heads.
      - mrow84 12y ago
        Somewhat farcically perpetuating the metaphor, is it not the case that Bill gets to know that Joe is interested in buying sheep once Joe has bought a few of them? And at that point Bill can outrun Joe and make money from his (very near) future purchases? That was my reading of the article - it seems that either there is a flow of information from the trading events to the fast traders, or the scenario portrayed in the article was very unlikely (though I guess that couldn't be ruled out given how much trading there is). noonespecial's metaphor seems to apply, what have I misunderstood?
        twoodfin 12y ago
        Somewhat farcically perpetuating the metaphor, is it not the case that Bill gets to know that Joe is interested in buying sheep once Joe has bought a few of them? He doesn't know with certainty. He can guess that's what Joe is doing, but he could be wrong and be stuck with sheep that he can't sell for the price he intends to ask. Every second he owns sheep is a second he's taking a risk that they'll go down in price, not up.
        mrow84 12y ago
        Yes, that makes sense. From what I understood of the article, and it really isn't my area so maybe I got something wrong, a buy order was not filled completely even though there were sell orders available to fill it. Doesn't that mean that people cancelled their orders whilst the buy was being filled, meaning that they must have been able to execute market actions out of order? That would put people who can act quickly at a big advantage on getting the price they wanted, because they can cancel and re-bid at a higher price. Sure it's a gamble, but it seems like one with little to lose. I've probably misunderstood something, but it certainly seems to be an issue that gets people very exercised. I presume there must be some competitive advantage in being fast, otherwise people wouldn't do it, so surely the only real issue is whether or not the consequence of exercising that advantage is socially advantageous?
simplemath 12y ago
Pretty sure he's suggesting adding a non-determistic lower bound to trade frequencies at network boundary

what does this do? (non-finance guy here)

by DevX101 12y ago

what does this do? (non-finance guy here)

The command introduces a delay of between 100 and 10ms to every packet. Screwing up trades that rely on being the fastest. http://www.linuxfoundation.org/collaborate/workgroups/networking/netem http://www.linuxfoundation.org/collaborate/workgroups/networ...

noonespecial 12y ago

Its a bit of a joke. Its a network thing, not a finance thing. What it does is adds a delay and more importantly, a bit more random delay to the time it takes the order to reach the exchanges server. Once you add in the non-determinism, HFT basically falls apart because you can't take your truckload of cash and buy yourself a place in a datacenter that's 2ms closer to the exchange and front-run everyone. It would be a wonderful thing to see all of those millions these guys have "invested" shaving a millisecond or two off their transaction times laid waste by a single command.

dzderic 12y ago
This "solution" will only make it harder for regular folks to execute orders, since HFTs will beat the randomness by shooting multiple orders through multiple order gateways.
- noonespecial 12y ago
  Guess we'll need a hierarchical token bucket with stochastic fairness queueing as well. We don't just need it to be random. We need there to be no way of ever quite knowing if any given order will beat another order to the exchange (within a given time period, of course). They won't know if they can beat joe ordinary, and they definitely won't know if they can beat the other HFT's. That might be enough to put a lid on it. Edit: For those playing along, here's the metaphor. Joe goes to market to buy sheep. Bill knows Joe is going so he sends a fast runner ahead of him to buy the cheapest sheep in town first so he can mark them up and sell them to Joe when he arrives. We try making everyone wait at the town gate for a random amount of time to give Joe a chance to arrive and get through. So Bill (being very rich) just sends 10 guys so one is very likely to be let in before Joe anyway. Next we introduce the stochastic filter. We make everyone line up and then shuffle the order every once in a while, but Bill still has more guys so he might still get one in first more often than not. Finally, we add the token bucket. For every one guy that we know employed by Bill admitted, we make the next one wait twice as long to get in, so if Joe and 10 Bills show up, Joe and the first Bill are essentially on even footing again because the 2nd through 10th Bill would have to wait too long to matter.
  - gd1 12y ago
    "For those playing along, here's the metaphor. Joe goes to market to buy sheep. Bill knows Joe is going so he sends a fast runner ahead of him to buy the... " Seriously, how many times have we discussed this issue on this site and we still get this bullshit. Bill doesn't know Joe is going. He doesn't. Get it through your thick heads.
    - mrow84 12y ago
      Somewhat farcically perpetuating the metaphor, is it not the case that Bill gets to know that Joe is interested in buying sheep once Joe has bought a few of them? And at that point Bill can outrun Joe and make money from his (very near) future purchases? That was my reading of the article - it seems that either there is a flow of information from the trading events to the fast traders, or the scenario portrayed in the article was very unlikely (though I guess that couldn't be ruled out given how much trading there is). noonespecial's metaphor seems to apply, what have I misunderstood?
      - twoodfin 12y ago
        Somewhat farcically perpetuating the metaphor, is it not the case that Bill gets to know that Joe is interested in buying sheep once Joe has bought a few of them? He doesn't know with certainty. He can guess that's what Joe is doing, but he could be wrong and be stuck with sheep that he can't sell for the price he intends to ask. Every second he owns sheep is a second he's taking a risk that they'll go down in price, not up.
        mrow84 12y ago
        Yes, that makes sense. From what I understood of the article, and it really isn't my area so maybe I got something wrong, a buy order was not filled completely even though there were sell orders available to fill it. Doesn't that mean that people cancelled their orders whilst the buy was being filled, meaning that they must have been able to execute market actions out of order? That would put people who can act quickly at a big advantage on getting the price they wanted, because they can cancel and re-bid at a higher price. Sure it's a gamble, but it seems like one with little to lose. I've probably misunderstood something, but it certainly seems to be an issue that gets people very exercised. I presume there must be some competitive advantage in being fast, otherwise people wouldn't do it, so surely the only real issue is whether or not the consequence of exercising that advantage is socially advantageous?

simplemath 12y ago

Pretty sure he's suggesting adding a non-determistic lower bound to trade frequencies at network boundary