5 ms·
This reminds me of how Google uses atomic clock and GPS for Spanner [1] Google: "“We can commit data at two different locations — say the West Coast [of the U
by Johnie 10y ago
This reminds me of how Google uses atomic clock and GPS for Spanner [1]
Google:
"“We can commit data at two different locations — say the West Coast [of the United States] and Europe — and still have some agreed upon ordering between them,” Fikes says, “So, if the West Coast write happens first and then the one in Europe happens, the whole system knows that — and there’s no possibility of them being viewed in a different order.”"
Renaissance Technology:
"Replete with schematic drawings, the filing describes a novel way for “executing synchronized trades in multiple exchanges.” The invention consists of not only sophisticated algorithms and a host of computer servers, but atomic clocks -- precisely calibrated to vibrations of irradiated cesium atoms -- to sync orders to within a few billionths of a second."
To translate what Renaissance is doing in technical parallel, they are trying to do a synchronous commit at multiple locations/exchanges at the same time. Submitting a trade to an exchange can be viewed similarly to committing data to a data center. By using atomic clock, synchronize these writes across multiple locations in effect eliminating HFT from jumping in.
If anyone is looking into prior art on this, Spanner is probably the closest I can think of. (I am not a patent attorney and don't want to turn this into a patent debate).
[1] http://www.theverge.com/2012/11/26/3692392/google-spanner-atomic-clocks-GPS http://www.theverge.com/2012/11/26/3692392/google-spanner-at...
- eru 10y agoThe whole scheme was already alluded to in Lamport's original paper. (Look for whenever he mentions `physical'.) http://research.microsoft.com/en-us/um/people/lamport/pubs/time-clocks.pdf http://research.microsoft.com/en-us/um/people/lamport/pubs/t...
- ovi256 10y agoGenerally speaking, as I'm not a patent lawyer either, the same technique applied to two different problem domains can generate two valid patents. The historical example is ship and automobile windscreen wiper.
- yandie 10y agoBut they're both in the software domain. Not to argue with you, but I don't see the problem any different from trying to synchronize two commits in a database. The same technique is not limited to Spanner or algorithmic trading, but other fields as well. It's not so much different from DHT or other algorithms, which have applications in multiple domains.
- toong 10y agoIsn't _software domain_ too broad ? You could argue ships and automobiles are in the same transportation-domain ?
- shiggerino 10y agoIn a sane world wiping liquids off of smooth, hard surfaces would be the same domain whether the material under whatever the wiper is attached to is in the liquid, solid or gaseous phase.
- dakr 10y agoIt's also the method by which signals from radio telescopes thousands of miles apart have been correlated after the fact at a central location. This has been the case for a couple of decades now. For instance, at the VLBA: https://en.wikipedia.org/wiki/Very_Long_Baseline_Array https://en.wikipedia.org/wiki/Very_Long_Baseline_Array
- amelius 10y agoWhat does that have to do with transactions? Analyzing radio signals is simple compared to running transactions, since it is only a one-directional signal flow.
- Bromskloss 10y ago> and still have some agreed upon ordering between them Funnily, that works only if the two observers are in rest relative to each other, as relativity theory tells us. :-)
- amelius 10y agoIn this case both observers are continuously accelerating towards the center of the Earth (because the Earth is spinning). So the theory actually doesn't work!
- Bromskloss 10y agoAren't we accelerating _away_ from the Earth to an even larger extent as we are held up by its surface, against the natural motion of falling downwards?
- joantune 10y agoI came here to say exactly the same about Spanner
- peter303 10y agoReminds me how Einstein derived special relativity. He was thinking about synchronizing clocks at railroad stations across the continent. The idea of simultaneity depends on the location of the observer. The traveltime of light crossing the globe in 1/14 a second matters little at train speeds. But thats an eternity for HFT and noticeable in a google search. These issue crop up on integrated circuits where clock cycles are shrinking and die sizes are increasing. Defining simultaneity where the microinstruction cycle time is much shorter than the chip or board propagation time is tricky. Each design has their methods to deal with this.