Slashdot Mirror


Google and Nasdaq Pursuing Nano-Second Precision In Network Time Protocol (nytimes.com)

"Computer scientists at Stanford University and Google have created technology that can track time down to 100 billionths of a second," reports The New York Times. "It could be just what Wall Street is looking for." Form the report: System engineers at Nasdaq, the New York-based stock exchange, recently began testing an algorithm and software that they hope can synchronize a giant network of computers with that nanosecond precision. They say they have built a prototype, and are in the process of deploying a bigger version. For an exchange like Nasdaq, such refinement is essential to accurately order the millions of stock trades that are placed on their computer systems every second. Ultimately, this is about money. With stock trading now dominated by computers that make buying and selling decisions and execute them with blazing speed, keeping that order also means protecting profits. So-called high frequency trading firms place trades in a fraction of a second, sometimes in a bet that they can move faster than bigger competitors.

42 of 203 comments (clear)

  1. wrong way by KiloByte · · Score: 4, Insightful

    Uhm, no. The right way would be to artificially add a random delay of several minutes, with a reproducible generator based on the orders' hashes and some seed known in advance, to avoid foul play. The generator would also need to be based on all previous orders, to avoid gaming the system if the seed is "accidentally" leaked. Details are more complex but it's all well-researched stuff.

    That would fix the high frequency trading abuse, which is nothing but pure theft, skimming a fraction from every bona-fide transaction.

    Just one of so many simple solutions... if only the high frequency traders wouldn't collude with the rule makers...

    --
    The creatures outside looked from Alt-Right to Antifa; but already it was impossible to say which was which.
    1. Re:wrong way by ShanghaiBill · · Score: 5, Insightful

      The right way would be to artificially add a random delay of several minutes, with a reproducible generator based on the orders' hashes and some seed known in advance, to avoid foul play.

      That would raise risk for market makers, increase transaction costs for small investors, and push even more big investors into off-shore dark pools.

      Before proposing silly "solutions", could you please explain what problem are you trying to solve?

    2. Re:wrong way by SNRatio · · Score: 2

      How about just putting all orders received in a certain amount of time (say 30 seconds) in the same bin. Orders within the bin are then matched to minimize the spread between bids and asks as much as possible. All of the orders that are successfully matched get executed simultaneously. All of the bids and asks, as well as successfully completed trades, then get revealed publicly simultaneously.

    3. Re:wrong way by ShanghaiBill · · Score: 2

      How about just putting all orders received in a certain amount of time (say 30 seconds) in the same bin.

      Because all the trades would be placed in the last microsecond of the time window.

    4. Re:wrong way by ShanghaiBill · · Score: 4, Informative

      It wouldn't matter when in that quantum the order was placed, everyone would be working from the results of the last quantum.

      But having a long time window and waiting till the end means you can incorporate more information. If there is a big block trade in London, and you can transmit the information quickly to NYC, you can place your order in the last microsecond of the 30 second window and screw anyone dumb enough to have placed their order earlier ... so everyone will wait. So your "time window" will do nothing to discourage HFT.

      What other effects will it have? It will increase transaction costs. High Frequency Traders are not investors, they are market makers. They find a willing buyer and a willing seller, arrange the transaction, and execute the trade. They make a profit on the spread between the buy price and the sell price. The problem is that once they locate the buyer and seller, they need to buy the stock from the seller first, then turn around and sell it to the buyer, but the buyer may have cancelled they transaction, or they may have already bought the stock from someone else, in which case the HFT is stuck with the stock and may have to sell it to someone else at a loss. If transactions are granulated to 30 second intervals, instead of say, millisecond intervals, then the risk of this happening is thirty thousand times higher , and the HFTs will insist on higher spreads, resulting in lower liquidity and higher transaction costs for both buyer and seller.

      Since the introduction of high frequency trading, transaction costs have fallen considerably, saving plenty of people a lot of money. The only losers are the old market makers that used to have lucrative sweetheart deals with the exchanges. Those old market makers are now bankrupt and gone. Good riddance.

    5. Re:wrong way by Wycliffe · · Score: 2, Insightful

      The problem is that once they locate the buyer and seller, they need to buy the stock from the seller first, then turn around and sell it to the buyer

      How is this a market maker? Why can't the buyer buy directly from the seller? Why do we need a middleman skimming off the transaction.
      Why can't the stock exchange computer do the matching? Take all the bids in a 30 second period, don't show them to anyone, match them up, and then publish the results. Matching buys to sells is what computers are good at. It is what HFT computers are doing except they are also skimming in the process. It would make much more sense for the stock exchange itself to be the market maker and do it fairly without skimming.

      Another option if you don't like the 30 second delay is to substantially increase the trading cost.
      Make it cost prohibited to do HFT. The average individual investor pays around $5 per a thousand shares
      compared to something like 2 cents per 1000 shares for HFT. Some liquidity is good but stability is better.
      HFT and day traders are leaches on the system. Transaction costs should be increased until the majority
      of trades are from long term and medium term investors.

    6. Re:wrong way by swb · · Score: 4, Insightful

      My gripe is that all this near-quantum physics in timekeeping doesn't really seem to be about fundamental economics, only about who or how gets to benefit from increasingly high speed trading. Too much of modern finance seems to be about how to manipulate the financial market itself to obtain profits vs. the actual fundamentals of business economics, i.e. the productivity of a given firm.

      I guess I'm not that worried about off-shore dark pools. The US and its dollar remain a major force in world economics not just because of the US domestic economy and in spite of the occasionally ugly behavior of the US government. Why? Because the US regulatory and legal system is mostly fair and mostly transparent, and it's judgements and rules are backed by the full force of the United States.

      No other nation can provide this and ultimately most capital will not choose to operate in an environment where the fairness and enforcement of contracts isn't guaranteed.

    7. Re:wrong way by Michael+Woodhams · · Score: 3, Insightful

      This is what I came here to say.

      The economic point of share markets is to allow companies to raise capital for long term investments. If doing stuff on millisecond or microsecond timescales is profitable, then that profit is doing nothing productive and can only be leeching value from actual productive economic activity. The investors who really matter economically are those who buy and sell over timescales of months or years, who will not be hurt in the slightest by trades being on a 30 second clock tick. (I really don't know what the time quantum should be - probably somewhere between 1 second and 1 hour.)

      --
      Quattuor res in hoc mundo sanctae sunt: libri, liberi, libertas et liberalitas.
    8. Re:wrong way by mlyle · · Score: 3, Insightful

      Yah. There's good reasons why a minutes-long delay is not great, because we want things to settle out relatively quickly... but 2.5-25 milliseconds of fuzz on the order book would be just fine.

    9. Re:wrong way by sjames · · Score: 2

      If the HSTs can buy from A and sell to B, the market already exists with or without them. Why not just A sell to B and they split the difference rather than giving it away?

      Perhaps it's time for the new market makers to go.

    10. Re:wrong way by phantomfive · · Score: 5, Insightful

      Before proposing silly "solutions", could you please explain what problem are you trying to solve?

      I don't know what problem he is trying to solve, but the problem that annoys me is front-running: listening for an order, then quickly buying the stock first (because you have faster connections), immediately turning around and selling it to the person who was going to buy it in the first place, but at a higher price. This does absolutely nothing to help the market as far as I can tell.

      --
      "First they came for the slanderers and i said nothing."
    11. Re:wrong way by Strider- · · Score: 2

      Why not just require a minimum hold time of, say, 30 seconds? That would pretty much put the HFTs out of business, and wouldn't significantly harm institutional and individual investors.

      --
      ...si hoc legere nimium eruditionis habes...
    12. Re:wrong way by religionofpeas · · Score: 2

      Why not just A sell to B and they split the difference rather than giving it away?

      Nothing is stopping A or B from doing just that, but if they don't take the effort to find the best price, somebody will trade both and make a small profit. That profit is a small price that A and B pay for the increased liquidity of the market.

    13. Re:wrong way by phantomfive · · Score: 4, Interesting

      No, limit means "buy at X price or lower." The front-runner drives the price up a few pennies, so without the front-runner you would have bought it at price Y < X, with the front runner you get it at a price Y + $.03 < X. So you're still losing pennies. It's a small enough amount that most people really don't care, but it's frustrating to get ripped off like that.

      --
      "First they came for the slanderers and i said nothing."
    14. Re:wrong way by phantomfive · · Score: 2

      Front running and other abuse of insider knowledge are not specifically related to high frequency trading.

      High frequency trading mainly profits by front-running. There was a book about this, Flash Boys.

      --
      "First they came for the slanderers and i said nothing."
    15. Re:wrong way by religionofpeas · · Score: 2

      High frequency trading mainly profits by front-running.

      My point still stands. If you don't want front-running, then target front-running. It's not that hard. Just make sure that an order from any trader stays secret until it hits the exchange where it becomes visible in the order book for everyone at the same time.

      If you just remove HFT, then front-running will still continue at a slower rate.

    16. Re:wrong way by ShanghaiBill · · Score: 2

      Why can't the buyer buy directly from the seller?

      If you are a seller, how are you going to find a buyer? Craigslist?

      Why can't the stock exchange computer do the matching?

      The exchanges are made up of their members. The members are brokerages that execute trades on behalf of their clients. You, as an individual, cannot log into the NYSE computer and execute your trade anymore than you could have walked into the pit during the old paper-based days. So why can't you trade with a "member" instead of a HFTer? Because the members are the HFTers. For all practical purposes, the HFTers are the exchange.

    17. Re:wrong way by deanpole · · Score: 2

      Front running is already illegal and strictly enforced in the United States. Brokers need to do manning which means giving fills to client orders whenever they match, even if the broker's order that executed was entered into the market before the client order was received. It might be 100 millisecond delayed to enjoy the benefit of hindsight, but clients get the best deal or the broker get fined and/or shut down.

    18. Re:wrong way by sjames · · Score: 2

      Nope yourself. Front running is when they speculatively buy and then cancel before it can go through if they can't find a buyer. This is the closely related but currently legal practice of finding a buyer and a seller and jumping between them using their faster system colocated in the same place as the exchange itself.

      HSTs don't hold stock. They don't offer liquidity. They don't "make" the market. They DO drive up costs and skim off the top.

    19. Re:wrong way by Harald+Paulsen · · Score: 2

      How would a 30 second minimum hold hurt regular investors?

      I'm a regular investor

      When I invest in stock I do it with the intention of holding it for days, weeks and months.

      --
      Harald
    20. Re:wrong way by religionofpeas · · Score: 2

      Why do you think they're willing to spend megabucks to shave off fractions of a microsecond?

      Because they make money doing it.

      Why do you think Amazon spends megabucks to make their order picking more efficient ? More profit for them, but also allows them to compete better, which benefits the consumer in the end.

      Same with arbitrage. Suppose you want to buy a stock in New York currently trading for $15.00, and for one brief second that same stock sells for $14.98 in Frankfurt. Do you want to spend your time with 5 open windows, fingers on the mouse, ready to cancel your order in New York, and quickly replace it with an order in Frankfurt (or any other exchange) ? Most people would agree that their time is better spend on something more productive.

      The HF trader will have systems automatically set up to detect the discrepancy. They'll buy the stock in Frankfurt for $14.98 and quickly sell it in New York for $14.99. You benefit because you can now buy it for $14.99. They make $0.01 in profit. You both win. The seller in Frankfurt also wins, because their order is processed quicker.

  2. It's all just enabling more bullshit by Zenin · · Score: 5, Insightful

    High frequency trading is entirely about subverting any remaining myth of the market or even less so-called "investing".

    What absolutely needs to happen is a flat transaction tax on any and all transactions, obliterate this entire train wreck of a financial vehicle from the entire economic equation. Simply out of basic fairness, why do I get charged 10% sales tax when buying a candy bar but not if it's a share of Apple?

    A simple 1% tax on transactions would overnight return the stock market to a system for investment rather than clever hacks to milk the real economy. If you don't think your stock is going to grow at least by 1%, you simply shouldn't buy it. An extremely modest 1% transaction tax would instill that sanity into the basic fabric of the marketplace.

    --
    My /. uid is better then your /. uid
    1. Re:It's all just enabling more bullshit by ShanghaiBill · · Score: 2

      The other option would be to make all markets call markets with 5 second rounds...

      That solves nothing. All trades would be placed in the last microsecond, and you would need still need extremely accurate time keeping to ensure an order doesn't slip into the wrong interval.

    2. Re:It's all just enabling more bullshit by Kjella · · Score: 3, Insightful

      An extremely modest 1% transaction tax

      A large low risk institutional investor like a retirement fund will average something like 3%/year. So every move would eat up 4 month's profit, that's extremely high and would slow liquidity to a crawl. If you wanted to kill HFT and reset the clock back to day traders then 0.01% would be enough, you could change horse every couple days but not buy and sell hundreds of times in a day like they do now.

      --
      Live today, because you never know what tomorrow brings
    3. Re:It's all just enabling more bullshit by Michael+Woodhams · · Score: 2

      Presumably bids would come labelled with the time slice they are intended to act in. Anything which comes late would be ignored, not applied to the next time slice (unless the bid's labels say this should happen.)

      --
      Quattuor res in hoc mundo sanctae sunt: libri, liberi, libertas et liberalitas.
  3. So when the time server fails the market crashes? by Bonker · · Score: 2

    So, in order to facilitate micro-trades, we're going to allow them to trade in terms of less than a second... which is disturbing all by itself when you think that even very fast internet destinations in the U.S. have latency on the order or 100ms or so. Even very slight congestion means that the financial exchanges have to simply trust the timestamps issued by the traders to get the prices and volume correct and cannot depend on the receive sequence.

    There's no room for malicious action there at all, and as we all know, the U.S. has absolutely no outside actors interested in manipulating events inside the country for their own ends.

    So ultimately, we're going to depend on some sort of time service run by Google and Nasdaq to validate those signatures. The problem there is that both those organizations have and will be aggressively targeted by bad apples the world over. TFA is a little light on the technical aspects of how, exactly, Google and/or Nasdaq is going to ensure that there's no forgery of timestamps. That service goes down... and Google and Nasdaq do have network problems from time to time... and suddenly all transactions are suspect? Heaven forfend that whatever algorithm used to synchronize and sign the transactions down to the nanosecond level is ever cracked. There's going to be a huge economic impetus for even very powerful countries to work on this. Imagine Chinese supercomputers manipulating American stock exchanges for their own benefit. I don't think that's too far fetched.

    And we're not just depending on Google or Nasdaq, but on everyone who takes part in issuing and signing these timestamps. Wow, that's a huge amount of attack surface.

    --
    The next Slashdot story will be ready soon, but subscribers can beat the rush and slashdot the links early!
  4. Re:So when the time server fails the market crashe by Mal-2 · · Score: 3, Insightful

    But it would be such a shame, and a blow to the Working Man, if High Frequency Traders can't operate! Think of all the bots that would be put out of work, you insensitive clod!

    --
    How is the Riemann zeta function like Trump rallies? Both have an endless number of trivial zeros.
  5. TSN by Jfetjunky · · Score: 2

    TSN Ethernet can already pull this off in most cases. IEEE 802.1... I've personally seen it do around 100ns sync or better depending on network configuration.

  6. Paper about the Huygens protocol by manu0601 · · Score: 4, Informative

    The New York Time article is mostly about how Nasdaq is eager to make more money.

    For the technically minded, refer to the paper about the introduced Huygens protocol for network time synchronization precise to the nanosecond.

  7. You need to convince voters that regulation works by rsilvergun · · Score: 5, Insightful

    take glass steagall. For 50 years it did a decent job of preventing the kinds of economic crashes we had in 2008. After 50 years without a major economic crash people came to the conclusion that since we hadn't had a blow up the regulations could go. Seriously, people think because bad things that regulations were passed don't happen anymore we don't need the regulations...

    I hate Ronald Reagan with a passion. He and his ilk (Karl Rove & the like) rolled back 100 years of hard fought workers rights in a few decades and turned the working class against the very notion of organizing for their own benefit. They turned "Union" into a dirty word and convinced folks to scrap the whole system because of a few bad apples and some Mafia interference. They were masters at it too. I'm singling him out because he was the spokesman for that crap. Mr "Government's the problem, not the solution". But crap like this is part of a broader trend to weaken the power structures that created and allow the continued existence of a middle class.

    --
    Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
  8. Re:Life in prison is a long time in nanoseconds by fisted · · Score: 3, Informative

    The irony is that while this improves NTP, "modern linux" uses "systemd-timesync" instead which kinda sorta knows how to get the time of day from one server with undefined tolerances, throwing away decades of work by intelligent people that went into making NTP what it is now.

    Who cares, right. What did those nerds know about clocks and time anyway.

  9. Re:Life in prison is a long time in nanoseconds by Anonymous Coward · · Score: 2, Funny

    I hope your weenie falls off.

  10. Re:GPS time by Anonymous Coward · · Score: 2, Interesting

    Getting precise time to better than a ns from GPS is not a trivial task (Hint: ionospheric weather routinely changes the naive answer by more than 10ns) but it is an achievable task. But at that level, everything matters. Is your antenna cable temperature-compensated? What's the temperature coefficient of your receiving hardware? Huygens seems to claim synchronization to the few 10s of ns, which is on the face of it a considerably easier proposition, but sounds interestingly challenging for commodity computers.

  11. Re:Life in prison is a long time in nanoseconds by Zontar+The+Mindless · · Score: 5, Funny

    Are you kidding me? Woohoo, yet another reason to hate systemd.

    --
    Il n'y a pas de Planet B.
  12. Re: Life in prison is a long time in nanoseconds by fisted · · Score: 2, Insightful

    timesync is the best tool because it barely does what you will settle for. Okay. Or are you trying to say that a more precise time will somehow harm your desktops?

  13. Re: Life in prison is a long time in nanoseconds by arth1 · · Score: 3, Interesting

    ntp is designed to minimize time jumps, skewing by small amounts more often. timesyncd is just a reinvention of old timed, including unfounded beliefs in predictable network latency, and, yes, clock jumps where none are needed.

  14. Glass Steagall got removed by rsilvergun · · Score: 2

    because people voted folks into office who repealed it. If congress was stacked with Liz Warrens and Bernie Sanders instead of Paul Ryans and Mitch McConnells it never would have been repealed.

    Getting rid of the bribes wouldn't hurt, but if the end result is still guys like Ryan, McConnell and other "Small Government" right wingers running the show then nothing changes. We need to convince people government can work so they'll stop voting people into office whose goal is to break government.

    --
    Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
  15. Few picosecond timing in use at smaller scale by joe_frisch · · Score: 2

    At SLAC and other large accelerator complexes we need to synchronize systems as the few-picosecond level (sub-picosecond for specialized systems), over a few kilometers. For us we have a near speed-of-light beam to allow a clear definition of synchronization (but the synchronization is maintained without the beam), and are now also installing a system that measures the round-trip physical time delay. The latter of course depends critically on the exact cable routing.

    For large scale (global) networks, its not obvious exactly what the goal is. Is the goal to ignore all cable delays? If so, I wonder how a computer's local time is certified to be synchronized at the nanosecond level - maybe some sort of GPS based synchronization where the computers position and GPS signal phase is used to determine time? (but what about the nanoseconds of GPS antenna cable?).

    It seems like defining what is wanted may be trickier than doing the engineering.

  16. Re: Life in prison is a long time in nanoseconds by Archtech · · Score: 2, Insightful

    timesync is the best tool because it barely does what you will settle for. Okay. Or are you trying to say that a more precise time will somehow harm your desktops?

    Do you demand that your car is capable of half the speed of light? (Although you will never drive it at more than 70 mph because you are a law-abiding citizen).

    --
    I am sure that there are many other solipsists out there.
  17. 1588 PTP by little1973 · · Score: 4, Informative

    There is already a standard with nanosecond precision (or better).
    It's called 1588 PTP.

    https://en.wikipedia.org/wiki/...

    --
    Government cannot make man richer, but it can make him poorer. - Ludwig von Mises
  18. Re:Because they're the ruling class by fafalone · · Score: 2

    What do you mean the other side isn't fighting? About half of them are fighting like hell to elect people who will screw them, the country, and the environment to help the rich get richer at the expense of everyone else. Because they've bought into lies about how much they'll personally benefit, or because of some wedge issue like abortion or guns or FUD about security, that politicians don't actually care about beyond them as a mechanism to keep people hating the other side (though both sides love using FUD to take away rights-- this comment isn't to suggest personal liberty is valued by either party).

  19. Re:You need to convince voters that regulation wor by Actually,+I+do+RTFA · · Score: 2

    I don't care if banks crash, as long as they don't take me with them.

    That's the entire point of Glass-Stegal. To prevent the banks that invest in whatever crazy risks they want to take from being the same as the banks that care if they fail. Like literally make them different entities.

    --
    Your ad here. Ask me how!