Slashdot Mirror
LSE Breaks World Record In Trade Speed With Linux
LingNoi writes with this excerpt from ComputerWorld UK: "The London Stock Exchange has said its new Linux-based system is delivering world record networking speed, with 126 microsecond trading times. The news comes ahead a major Linux-based switchover in twelve days, during which the open source system will replace Microsoft .Net technology on the group's main stock exchange. The LSE had long been criticised on speed and reliability, grappling with trading speeds of several hundred microseconds. The 126 microsecond speed is 'twice as fast' as its main international competitors, the London Stock Exchange said. BATS Europe and Chi-X, two dedicated electronic rivals to the LSE, are reported to have an average latency of 250 and 175 microseconds respectively. Neither company immediately provided details. But many of the LSE's older and more traditional rivals offer speeds of around 300 to 400 microseconds. Nevertheless, Linux is now standard in many exchanges, including the New York Stock Exchange."
Can someone explain to me what the system does in these 126 microsecond? Is it sending packets through the world, doing some complicated calculations, solving locking resources? It seems an awefully long time to add to a table and update some stats.
NB: The message above might reflect my opinion right now, but not necessarily tomorrow or next year.
Trading this fast brings the market closer to optimal economic efficiency, where prices at any instant accurately reflect value. Latency contributes to the very inefficiencies that you blame these "large investment firms" from profiting off of. These high-speed arbitrage and "quant" investors may make a profit without creating a product (though collectively their track record is not very good financially), but their profit margins are vanishingly small and they serve a critical role in equalizing prices between markets and keeping prices up to date as market conditions change. In short, your complaint about these trading practices smacks of jealousy and sour grapes, and it ignores the valuable role they serve in the markets.
I am a geek attorney, but not your geek attorney unless you've already retained me. This is not legal advice.
Okay, well to anyone who's ever had to work with Accenture code, it would tell them a lot.
I agree wholeheartedly that it has nothing to do with MS vs Linux, I think it has to do with another shoddy Accenture implementation. Even the .NET decision has nothing to do with it IMO, I'm a firm believer that algorithms and design have far more impact than OS or language choice.
Oh, and calm the hell down. It's a discussion, not a flamewar.
Co-location. A lot of exchanges will let you put your servers in their building. Then the arguments start about who's network cable is shortest and which router it's on. I'm not joking.
Yet oddly, in the UK at least, there is a tax on actually buying shares, so my long-term investment account has various charges for stamp duty as my money goes in there. The tax is only applied to buying actual shares. It isn't applied when messing around with futures and complex derivatives etc that have caused the problems.
I think this needs to be exactly the opposite way around. I don't think there should be a tax for dealing in shares, but instead a tax on dealing with various futures products. Not too severe a tax but enough to dissuade this sort of high speed speculation.
It's a great server OS, sure, but lets look at this realistically:
- the Windows / .NET trading system was based on Windows 2003 and SQL 2000, and was deployed in 2005.
- the Linux-based system is under development now, to be deployed next year.
You missed
- the Windows / .NET trading hardware has been upgraded continuously because it was unable to cope with the load.
Just based on that, you'd expect substantial performance differences from just using newer hardware.
Sure, except for the part that the both are running on new hardware.
Chances are that the original kit was certified as a part of the solution, and hasn't been replaced since.
"Chances are" - except that is 100% wrong. They had problems since day 1, which were blamed on the hardware, so they've been constantly upgrading it trying to fix the problem.
Even ignoring the hardware and the OS, one would expect 90% of the performance to be determined by the application, not the OS. Decisions like writing the software in .NET versus C or Java, or using a special-purpose Java runtime would make a huge difference, irrespective of the OS.
The old system was written with the help of MS. They were the ones that said that .NET was the best way to implement it, and they even touted this in their press releases.
On top of this, the software stack is completely different, and developed by a different team. Just about every design decision, small and large, will be different.
Of course it's completely different - that's the entire fscking point.