The importance of 'ping' - latency/connection speed in trading.

Discussion in 'Order Execution' started by gotmessner, Sep 8, 2005.

  1. def

    def

    i can't speak for liffe but it's not done that way elsewhere.
     
    #11     Sep 11, 2005
  2. can anything be done, on the client side, to decrease ping time?
     
    #12     Sep 14, 2005
  3. tomcole

    tomcole

    Try this website - http://www2.dslreports.com/

    Then go buy Cisco routers, high end modems and ask your ISP about guaranteeing service times.
     
    #13     Sep 14, 2005
  4. JackR

    JackR

    Yes, but rather expensive.

    I've not had any reason to look into this kind of stuff for a few years but I'd say that the info that follows would still be valid.

    You need a leased line to your broker's server. As part of the lease you may specify a maximum latency. The laws of physics (speed of light/electron flow) will determine the absolute minimum latency. The routing used by your service provider to get from your equipment to the far end is the next limitation. The latency requirement (average and peak) will determine how your provider engineers the route.

    Leased lines are priced by capacity (64 kbps, 128 kbps, ....1.544 Mbps, etc.), straight line distance between the two ends, latency (if non-standard) and level of service (reliability - do they need to provide alternate routing to the customer premises - VERY expensive).

    In general, I'd say the cost would be prohibitive for most at-home traders trading for themselves.

    Jack
     
    #14     Sep 14, 2005
  5. JackR

    JackR

    Be cautious using the term "service time" with a telecom firm. This normally refers to outages. That is service is available 99.99% of the time, 99.999%, etc.. It has nothing to do with latency (network delay).
     
    #15     Sep 14, 2005
  6. banana

    banana

    Use a shorter cable.
     
    #16     Sep 14, 2005
  7. Probably not. One thing you could do to check the performance of the client side (computer router/modem combination) is to ping or traceroute your ISP ie a mail server or time server at the ISP.

    On my machine I get about .5 msec to the router/modem, about 10 msec from modem to first external router and about 22 msec total (five hops) from the ISP's servers which are located 650 miles away.

    Seems very acceptable to me. I very much doubt that replacing the cheap Netgear modem/router with expensive Cisco kit would make any practical difference.

    The reality is that the data rates of ADSL or cable are pretty small beer in the larger scheme of things and low end kit should easily deal with them.
     
    #17     Sep 14, 2005
  8. Here is the reply from Euronect.liffe SVP US Technology:


    >>
    Not all exchanges do what LIFFE does. A few things differentiate LIFFE CONNECT(R) from other systems.

    1) Multicast transmission of market data.

    When an change occurs in the market (price/size), a multicast message is sent out via the API to all directly connected end-users. This message is sent out once and subsequently "heard" by all participants at the same time however with slight variance depending on where in the world the end user resides. It is pure physics in that a message sent from London to Paris arrives sooner than a message sent from London to Chicago. That being said, no other transmission method is as "fair" as one sent via multicast.

    In contract, a unicast sent message requires separate sessions with each individual end-user, which then follows that any one user could receive a market data message sooner or later than another end-user.

    2) Bandwidth Management

    In the event market data sent from the host to the end user consumes the available bandwidth at the end-user level, LIFFE CONNECT(R) will "throttle" the data for all market participants, ensuring that no one end-user is receiving information that is not available to other end-users. Other systems typically employ a per end-user queue, whereby if bandwidth for an individual member reaches capacity, that one user will be queuing messages while other market participants (who may have larger amounts of bandwidth) would be receiving messages.

    3) Dynamic release of market data

    LIFFE CONNECT(R) disseminates all market data as it becomes available, rather than "pulsing" or aggregating market data every quarter second.

    4) Full depth of book

    LIFFE CONNECT(R) shows all resting bids and offers above and below the best bid and offer.

    As far and I am aware, CME applies (1) and (3) above, Eurex (1) (4) and LIFFE/CBOT all of the above.

    Now to answer the below question specifically [edit: I asked to comment Fredbloggs post], there is no speed of light leveling employed on the network. Aside from the fairness components described above, we do not delay data to Paris so that it is in line with data arriving in Chicago a few milliseconds later. To do so is difficult, and relatively unfair in attempting to objectively determine what metric should be applied to the Paris delay. If immediacy is of great concern to a firm, it would be best for them to set up shop next to the host. However, I have yet to see a firm that cannot trade profitably from the US because they experience a 40ms difference in immediacy.
    >>
     
    #18     Sep 15, 2005
  9. FredBloggs

    FredBloggs Guest

    lol - well so much for my mates version!!

    just goes to show we shouldnt believe everything we hear!

    cheers bernard.
     
    #19     Sep 15, 2005
  10. I'm too old to be a scalper, so maybe what I have to say doesn't apply.

    But it seems to me that if a trader can claim, "I'd be profitable if only someone didn't keep beating me to the fill by a quarter second", he's playing the wrong game. After all, it takes an average .70 seconds (700 ms for you math impared) to execute a voluntary motor response.
     
    #20     Sep 15, 2005