Hacker News Reader

Comment by rsync | Hacker News Reader

"HFT firms will be the first customers, seeking a few-millisecond reduction in the transmission of realtime data between markets. Starlink has already said as much, and I think they rely on this as their first and primary source of revenue."

I am not sure this is a good prediction ... the shaving of milliseconds that has been pursued by HFT firms (and other such actors) has progressed to the point where they are measuring cable lengths inside the datacenter.

Even at a greatly reduced orbit, I can't believe that bouncing up and down via satellite as part of your link could be any part of a time-arbitrage recipe in 2020 and beyond ...

View on HN

> I am not sure this is a good prediction ... the shaving of milliseconds that has been pursued by HFT firms (and other such actors) has progressed to the point where they are measuring cable lengths inside the datacenter.

You're conflating optimisation of latency within a colo - how quickly can you send an order in response to seeing an event on that exchange - with latency between colos - how quickly can you send an order in response to seeing an event on another exchange.

Within a colo, latency is measured in microseconds and fractions of a microsecond. Between colos, it's milliseconds. Some random source here has 4.13 ms for New York - Chicago:

https://www.motherjones.com/kevin-drum/2013/11/final-frontie...

At the moment, to play that game, you need to build a microwave network. Someone needs to build towers in a line between your two exchanges - buy the land, get planning permission, placate the locals, build the tower, install microwave antennas, hope the weather is okay. That's expensive!

With Starlink, there will be a reasonably direct path between any two points on earth - on demand! The path will be longer than a dedicated microwave chain, although some of it will be in vacuum, which will save some time, but it will cost the end user no money and no time to build. If you come up with a trade idea that needs a low-latency path from Moscow to Stockholm, you can send a request to the Starlink API and just get one in moments. You can be up and running with your trade months or years before a tower-building competitor even joins the game.

The thing is people have already built most of those microwave tower paths here on earth and in addition SpaceX probably isn't going to set it up as a mesh where intra-Starlink traffic stays goes directly. And even if they do you have to go up then across (probably in a zig zag) then back down which will always be higher latency than the point to point microwave distance.

>people have already built most of those microwave tower paths here on earth

I can assure you that lucrative paths like New York-London and New York-Tokyo have not been built yet. In fact, the vast majority of lucrative trading paths have not been built yet due to geographical restrictions. Yes, fiber optic links exist, but those only go up to 0.7c.

To a trading firm, the choice between a satellite link that does ~0.9c and a fiber optic link that does <0.7c is obvious.

That all depends on Starlink supporting point to point inside their network. Also that path will be more zig-zaggy than the great circle fiber path and importantly will vary during the day as the satellite tracks shift relative to earth. [0]

[0] Unless their in a harmonic orbit I guess not 100% sure abou that one. There should always be some jitter in the connection time just from the satellites moving even if their ground tracks are very stable.

I don't know enough about their infrastructure, but presumably SpaceX could offer that as a premium service? "Most direct" path for your data at a much higher dollar-per-MB cost vs the mesh offering?

4.13ms is one way, fwiw, RTT is 8.3ms or so.

> Some random source here has 4.13 ms for New York - Chicago

I (sitting in the Bay Area, at home, with Sonic fiber) am getting 5.122ms RTT ping times to MIT's CSAIL server. That's way further than NY-CHI.

Edit: oops, I forgot basic physics. 1ms =~ 300km. It should take at least ~10ms each way. I must be hitting a CDN.

That is physically impossible (min RTT is ~40 ms[1]). You are probably hitting a CDN.

[1] https://www.wolframalpha.com/input/?i=San+Francisco+to+Bosto...

A couple things:

1. Signals travel faster through the air and vacuum of space than they do fiber optic cables or electrical cables. 2. If there is an HFT trader based in the U.S. looking to trade based on events happening in Spain or Japan, they still want to get data across continents in the fastest way possible.

EDIT: Apparently other posters traded on the information arbitrage opportunity of replying to your post several seconds faster than I did. gg.

Traveling twice as fast three times as far isn't a win.

People already gave pretty exact distances in other posts.

>Even at a greatly reduced orbit, I can't believe that bouncing up and down via satellite as part of your link could be any part of a time-arbitrage recipe in 2020 and beyond ...

Really? Just some rough napkin math here, but only photonic bandgap fiber reaches near light speed (~99.7%) and I'm not aware of any large real world deployments of that or even near future ones. Regular fiber is only around 70% c. Transmissions through air (and obviously near vacuum as it exits the main atmosphere) are very nearly c. So from a latency perspective Starlink should in principle have up to a 42% inherent raw transmission speed advantage right? So that then has to go against the extra RTT to get up to a satellite and back. It looks like the lowest V-band shell is planned to be at just 210 miles, with another shell at 340 mi, and a final Ku/Ka-band shell at 710 mi. I don't know if the lowest shells are supposed to have any point to point mesh capability or must go to a higher shell first, but if they can that's only about a 420 mi inherent RTT penalty. For a link over 1400 miles, that seems like it might give them an advantage. Of course, there will be other routing factors, but then again on the fiber side only a few links (NYC to London for example) are actually really point to point optimized, most routes around greatly separated parts of the world have their own extra distance penalties due to transmissions having to take the legs of triangles rather then the hypotenuse.

I'm not an HFT person, but I assume there is probably arbitrage to be done between, say, NYC/London and Tokyo or other Asian exchanges and the like too. A lot of those are 6000 to 7000 mi, something like NYC to Sydney (ASX) approaches 10,000 mi. At those kinds of distances even if the signal has to go all the way up to the outer shell to mesh it looks to me like it could still beat a conventional fiber transmission right? I don't know if HFT firms will be that core as customers or how important they are in the scheme of things here, but a rough check certain seems to indicate there should be potential there in terms of the physics and what infrastructure is current deployed.

Light in fibers travels at around 2/3rds of the speed of light, while in air or vacuum it travels at (approximately) the speed of light. So a path through space can be 50% longer, and your signal will still arrive at the same time.

Consider that the distance between New York and London is ~5500km, and these satellites are only ~500km up, that gives you quite a bit of margin to be faster.

The same argument is true of WiFi over a network cable. Depends on the packet loss on the way down. You also potentially might have to send more data for security reasons too.

Also, if the satellite is halfway between them (without taking the curvature of the earth into account, assuming a flat earth and the signal only goes through one satellite), the signal still has to travel 5590 km to cross the Atlantic (at almost c). Probably a solvable problem but not a trivial one.

It might be if you're arbitraging markets against each other. Starlink orbits are at 340 km, i.e. 1.1 ms altitude; if you're sending data between London and New York, you can afford 2.2 ms of uplink/downlink if it allows your signals to travel at c instead of 0.7c.

Will Starlink actually transmit intra-Starlink traffic point to point like that is a big question.

Consult the original source: http://nrg.cs.ucl.ac.uk/mjh/starlink-draft.pdf

These first satellites don't have the free-space lasers I modelled in that paper, but my simulations show they can still beat fibre on many paths without inter-satellite links, by relaying via groundstations:

https://youtu.be/m05abdGSOxY

Later satellites will have inter-satellite links, but probably not till near the end of this year, and most of the first phase will already be deployed by then.

In many places for HFT the comparison is against microwave links though not standard fiber.

Indeed, and I mention microwave links in the paper linked above. But those are relatively short distance: New York - Chicago, London-Frankfurt, and similar. Where Starlink and Kuiper can provide real latency benefits is on long distance paths, trans-oceanic, or passing over countries where it's geographically or politically infeasible to built microwave links.

Reply on news.ycombinator.com