ARCA Cancel - How long does it take?

Discussion in 'Order Execution' started by WinstonTJ, Jul 27, 2010.

  1. Anyone know how long it takes for ARCA to turn around a confirmation of cancel and send the message out?

    I know my latency, I'm asking about the time it takes from when ARCA receives my CXL to the time they send me my OUT.

  2. Depends on FIX matcher, 3.x, or 4.x.

    The latest version supports a new C/R type that is considerably faster.

    Only way I know of to get real info is to send some orders, call them up, and ask for an audit trail.

    In my experience the FIX engine is probably 2.5ms, and the latest ARCA Direct is about 1ms.
  3. I thought it was -1ms.

    You are posting too soon after your last post. Try again in a few minutes. What do you think this is, a high frequency posting board?
  4. Well, not that I trade on ARCA frequently (not much of an equities person), but I distinctively remember a chat I had with a firm that is active in HFT strategies on ARCA, and they indicate the "raw" ARCA response time is something like 0.6 to 0.8 ms.

    Modelling the response time, tack on small network latency, message queueing time, i would make educated guess the actual "cancel accept" latency is around 2.5 - < 3 ms.

    Heh, I am going to be at a conference with a few Getco ppl, I should just Ask. Heh.
  5. When was that? 4.x (I can't remember) is faster. They still suck relative to Nasdaq and BATS, though.

    And if raw is 0.75ms, total time should be under 1ms if the logic is colocated.
  6. I've been measuring and getting my audits - I have too many hops I think. Its taking me about 150ms to get a fill and 250ms to place an order and then cancel that order and get the out confirm.

    I've been using ARCA because when I take I find that generally they have a better smart order router - but perhaps its time to try other ECNs. I need to pass through ML/PAX, maybe I should try BofA?

    I'd love to get down to sub 25ms but I'm not colo. I do have racks and gigabit ethernet, etc. but I'm over internet not FIX and I'm not in the datacenter where the nyse execution boxes are. How much will that improve speed - the trades I'm considering this for probably don't make enough to support that type of hardware upgrade/improvement.

    Do you guys know what is reasonable/average for orders over fast internet?

    Appreciate the replies so far.
  7. Do you have anything positive to contribute?
  8. it depends on where you're located, but from the numbers you're quoting sounds like west coast, and even then sounds slow. more than likely your broker is introducing a good percentage of that delay as well.

    advice: if your strategies are time sensitive, then solve this problem by getting properly colo'd; otherwise, don't spend too much time on it.

    there's very little in terms of consistency or repeatability in latency over the internet, and trying to shave time here or there really makes no difference if every other day 'hop3's' buffers overrun and you're looking at 5x the average.
  9. Additionally, if you care about speed you want to avoid the exchanges routing your orders. I have stuff that necessarily does that and when it's busy they can take 3+ seconds.

    Unfortunately the fragmented equity marketplace is making it difficult to do this stuff well cheaply.

    In any case, 250ms doesn't make sense. That's going through some serious packaging / checking / long-way-around routing. To give you a data point, the fastest available line from Chi <-> NJ is ~ 13ms roundtrip (I'm cheating a bit, it actually lights for production trading in a couple weeks).

    No idea what's in the Pax Order Router these days, so can't help there.
  10. Just figured out a bunch of things. I'm in NYC, <1ms between my boxes and the execution server (Sterling). All the delay seems to be associated with firm limitations, risk, etc.

    I'm looking at a different execution platform that will be faster. I need to execute in ~50ms (from NYC to NYC over gigabit internet). Any suggestions would be welcome.

    #10     Jul 28, 2010