Slashdot Mirror
Heat and Humidity Slow Down High-Frequency Trading Due To Microwave Links (hackaday.com)
szczys writes: Even tiny slowdowns have major ramifications on automated stock trading. To put the computing power as close to the markets as possible, microwave links (point-to-point links via dedicated microwave dishes) connect Wall Street to server installations in New Jersey. Hot weather, especially when accompanied by high humidity, slows those links down enough to make an impact on trading. From a report via Hackaday: "For short-haul links around the financial centers in New York, though, dedicated network links are favored for low-latency connections. Rather than trusting their trades to the vagaries of the internet and risk an unfavorable routing path or a cable severed by an errant backhoe, high-frequency trading firms often rely on microwave links to exchange information. [...] As it turns out, those microwave connections are the weak link in the system. During the early July heatwave, the links were experiencing slight delays in transmission times over that 16-mile path and throwing off the timing of the trading algorithms. The delay was minuscule -- on the order of 10 microseconds -- but in a business where millions are made and lost in seconds, that's substantial." Last month, Bloomberg reported that high humidity was impeding radio transmissions among three New Jersey data centers where U.S. stocks trade. According to a note Nasdaq sent customers, it took about 8 microseconds longer to send info from the stock exchange's facility in Carteret to the New York Stock Exchange data center in Mahwah, and an extra 2 microseconds to send data to Cboe Global Markets' exchange in Secaucus.
There. I said it.
Instead of not. Sovled. Solved.
Hah, see what your microwave link got you?
When I sell shares in my Fidelity account, the brokerage holds my money for 3-5 days.
Kiss my ass Wall Street. I bet you're using my money even in this high frequency trading shit.
I'd even wager you blame it on "the law" which probably says you have to settle trades within X hours, so you make sure you hold my money exactly X hours, even though it was undoubtedly available nearly instantly.
TSLA longs just lost $1.05 Billion dollars today.
I doubt it.
Maybe if fewer quants tried to hedge things to change a method for investing capital into a method for legalized gambling, the world would be a better place.
(caveat - some of my cousins work for such firms)
-- Tigger warning: This post may contain tiggers! --
A reason for the Republicans to combat Global Climate change.
I would not shed a tear if someone jammed the link and sunk the company.
I used to work tech in and HFT firm. They can suck a dick. They're all just jumping over each other to out cheat each other anyways.
Here's hoping they get another 10us added on
couldn't resist...
nothing to see here - move along
The delay was minuscule -- on the order of 10 microseconds -- but in a business where millions are made and lost in seconds, that's substantial."
My sympathies to all the high-frequency traders who have to wait an extra 10 mu to rip the rest of us off...
It must have been something you assimilated. . . .
Any kiddos in NJ have a drone and a sheet of aluminum foil?
/Nelson
I am Slashdot. Are you Slashdot as well?
I hope some person / voice of conscience at the SEC or Treasury is advocating to put a per-transaction fee on stock trades in the long run someday. Even 1/10000 of a cent. That probably would be enough.
I find it hard to understand how microsecond quantitative trading that screws all of society by skimming ~ a penny off of every 100 shares, is providing any extra liquidity that benefits the market (which is their claim to why this is a valid thing to do). We're all being played by these quantitative traders, for the pure benefit of billionaires.
I doubt it.
Maybe if fewer quants tried to hedge things to change a method for investing capital into a method for legalized gambling, the world would be a better place.
(caveat - some of my cousins work for such firms)
Of course not. By it's nature it's an activity that produces nothing (it's not like a company's going to shut down if someone doesn't buy its stock for that particular tenth of a second), punishes investors (it raises the price for the person who wants to invest in the company but can't afford the millions in infrastructure cost to get a give few seconds advantage), increases the divide between the rich and the poor (see #2), and is a drain on human society.
From a geeky perspective it's a fun idea, though. It's just using technology for pseudo-evil that happens to make you rich.
Somebody call the Whaaaaaaambulance!
This is terrible, because we all know that high-frequency trading is totally necessary for mankind to thrive.
fetch my fainting couch at once!
If summer humidity is having an effect, what happens on a rainy day when humidity is 100%?
Also, does this mean someone could disrupt these transmissions with something like a drone...?
Too bad 10ms is too short for HFT promoters to think about how toxic their practice is to society.
Jeez these people are idiots
I have crappy cell coverage as my house as it is, but this Summer it is nearly non existent. Have had to go drive down the road a couple of miles a few times just to get a signal able to stay connected. Can't wait for cooler weather so I can get at least a single bar again.
It's interesting that one motivation for using microwave communications is to avoid the risk of disruptions like inadvertent cable cuts. However, cables buried in the ground are probably more resilient to attacks than 16-mile communications over open air. If humidity spikes can impact communications, how about steam chimneys, kites, and balloons along the path placed by competitors, not to mention intermittent random jamming.
Follow the money, and if you can't follow that, follow the excuses to those who claim to have lost the money.
Yes, in high winds microwave transceivers do have loss of signal (LoS) and other issues. In high humidity and temperature there are other factors. All that is true.
However, a 16 microsecond latency is UNDETECTABLE and IRRELEVANT. To put it in perspective, the latency of a 1500 octet Ethernet frame over a 1Gbps LAN including processing by the transmitter and receiver running a Real-Tiime OS (which none of these Windows-using traders run) is 15-30ms which is 1000x slower than the 16 microseconds described in the article.
If your underlying network hardware and software stack can't process the data in less than 15ms (best case) then your extra 16 microsecond delay is UNDETECTABLE AND IRRELEVANT.
Caps for emphasis not for yelling. This is just traders making up excuses for why they lost money and suck.
E
running a Real-Tiime OS (which none of these Windows-using traders run)
Windows is only the HMI. The lower level network stacks and trading apps are running on highly customized and stripped down networking stacks. Linux was hot some years ago. But I imagine that some trading systems have gone to FPGAs/ASICs. So a 16 microsecond latency represents thousands of instructions on some of the more mainstream architectures. More on the really high performance stuff.
Have gnu, will travel.
The climate crisis is nigh. Never mind the fires and drowning polar bears. It's affecting the bottom line of rich people. Put up those windmills and solare panels now!
Have gnu, will travel.
I think you're confusing microseconds and milliseconds.
Milliseconds (typically abbreviated ms) is noticeable. Miroseconds (typically abbveriated us) is not. 1 ms = 1000us.
Put another way 16us=0.0016ms. That's like an eighth of a millisecond. 1/8ms. Not noticeable.
It doesn't matter how fast the FPGA, CPU, ASICs (listed in increasing order of performance) can process data if the underlying network hardware is 1000 times slower.
E
However, a 16 microsecond latency is UNDETECTABLE and IRRELEVANT. To put it in perspective, the latency of a 1500 octet Ethernet frame over a 1Gbps LAN including processing by the transmitter and receiver running a Real-Tiime OS (which none of these Windows-using traders run) is 15-30ms which is 1000x slower than the 16 microseconds described in the article.
You make several (wrong) assumptions.
Their goal is to get from their processing center into the exchange as quickly as possible. If they are first to arrive at the exchange network then they are most likely to be routed through that network first and be processed first.
You don't need Ethernet to do that. ATM with its 53 octet cell size is well suited to the task, or even a custom on the wire protocol that's converted at the exchange pairing point. Even if the exchange is using TCP/IP over Ethernet you just need to be first onto that network and the earlier you get there than the competition the more likely you'll remain first through it until your trade is processed.
You don't use Windows (or even Linux) on X86 when you want supreme low latency. You use a RTOS. I have seen job ads for "investment companies" that want RTOS, and all sorts of obscure network technology experience.
If you want less latency you do it in hardware as far as possible. I recall seeing somewhere that the HFT crowd were dabbling in FPGA and custom silicon to do just that.
Remember, these people have virtually unlimited money to shower on solving their problems because their investment pays off 100-fold. It's too bad all their research and solutions are closely guarded secrets - if they were published they might release a total gem that changes the computing/telecomms industry.
In the wireless world LoS = line of sight.
It does not mean loss of signal.
I'm on a WISP (Wireless Internet Service Provider) and it is definitely slightly slower when it's hot.
Ferret
Sic gorgiamus allos subjectatos nunc
Put another way 16us=0.0016ms
Good thing we all went metric because dividing by 1000 is easy.
16 microseconds is 0.016 milliseconds.
That's like an eighth of a millisecond. 1/8ms. Not noticeable.
To you or me, no. But to specialized HFT trading systems with optimized hardware and networking stacks, that is ages. Particularly if you are trying to beat some other HFT system to a trade.
Have gnu, will travel.
Only liquid water impedes microwaves. Vapor doesn't, nor does ice and snow. Even then, liquid water only marginally affects frequencies above 2 GHz. It's a big problem above 11 GHz in the Ku- and Ka-bands as satellite TV and satellite internet users are well aware.
Terrestrial point-to-point microwave is a little above 3 GHz, which has been demonstrated to be unaffected at all by liquid water, and completely unaffected by humidity. The problem is that this article doesn't even bother to mention the frequency used.
Humidity doesn't affect anything. The problem, from my humble perspective, would be processor throttling due to high heat at the receiving site. More so, the BER, or Bit Error Rate, is not even mentioned in the article even though it's the chief factor used to judge how a digital link functions.
Kriston
Also, you're all wrong. (Well, most of you are.) If your computer is the best you can afford, and your algorithm for deciding on whether or not and when to to make a trade is as efficient as possible, and your bid or order or whatever is less than 1 microsecond BEHIND the one that it could have beaten if your signal had managed to be only 1 microsecond faster, that can make a VERY big difference if your order was to buy all the shares of this one stock, and the one that beat you said, "buy all the share of this one stock," you go from being able to buy all the shares of this one stock, to being able to buy NONE of them. Or something like that.
Inside the exchange itself, they let traders put their computers inside the actual exchange, and the cable-runs provided are all carefully measured so that no matter how far a given computer is from the one that handles the requests, physically, they are all the exact same distance in TIME, meaning the time it takes for the signal to go from the back of your computer to the other endÂis exactly the same for all the others, so it's fair. Timing matters THAT much.
If you get beaten by only 1 microsecond, that microsecond might as well be an eternity. Ask anyone who's ever lost by a nose in a photo-finish, or anyone who had money on the loser.
Here's more on this. https://www.wnycstudios.org/story/267124-speed/
(There's multiple segments of this podcast: the first is about the Pitch Drop Experiment, not really relevant here, but about 24 minutes in, they talk about HFT.)
Our reign has gone on long enough. Indeed. Summon the meteors.
Anecdote: Somebody told me they were building the packets on the fly before they knew they even wanted them sent, and if a decision was made that the packet shouldn't be sent then they'd fudge the checksum at the end of the pipeline.
Or to paraphrase somebody who wouldn't add more flash memory to a device because it costed 2 cents more per unit: when you add volume it ads up to millions.
"Everybody's naked underneath" -- The Doctor
So if you can figure out the line of site for a competing company can you put a very nice fountain in the middle? Or you could accidentally fly a drone with a copper mesh banner behind it advertising your company in and out of the line of site.
The way banks make money on these millisecond information advantages is by learning in advance about pending orders big enough to move the market, and trading on that knowledge before the big order comes through. This is called "front running", and used to be illegal. The law hasn't changed, it just isn't enforced anymore.
https://www.youtube.com/watch?v=6YN8pSaWggQ
Correct me if I'm wrong, but it seems like this would only be a problem for high frequency traders that don't pay to colocate at the New York Stock Exchange.
For those that don't know, you can pay to have a server on location at the NYSE site in New Jersey. The exchange goes through great lengths to make sure all clients have the same latency. I believe all of the network connections are equal length.
One of our competitors trademarked the term "hypothesis". From now on, we will call them "boneheaded ideas".
This is what I've always said about the weakness in HFT:
1). What HFT offers is speed. OK, what if a competitor offers (some ridiculously small increment) faster speed. They take your business and you are toast;
2). What beyond mere speed, does HFT offer? Where's the value-add? When it comes down to it HFT is just tech-enabled arbitrage. There's no sophisticated value-added to justify much of a business.
And no, don't bother lecturing me about DC co-location, your fancy optical waveguide/lasers/high end servers/low latency TCP/IP stack/speed enhanced applications. All that is just a desperate attempt to keep ahead of the sharks nipping at your heels.
One tiny misstep and your business disappears. Like with this humidity-driven microwave thing. Really, you can be taken out by humidity? Something that happens every summer?!
You know what else might happen? A competitor buys better access than you have. Again, poof goes your business. Talk about a house of cards.