Submitted by “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 record breaking times were measured on the LSE’s Turquoise smaller dark pool trading venue, where trades are conducted anonymously. That network switched over to Linux from Cinnober technology two weeks ago. Speed is crucial as more firms trade automatically at lightning speed, using advanced algorithms.”
Well, if (as some believe) the Linux desktop is dead, then long live the Linux server!
Migrations like this are *great* to see. All the more so for being in such a high-profile “white-collar” environment where things Must Work(tm).
Interesting to see that the article mentions that “Linux is now standard in many exchanges.”
Given that Linux will be replacing Microsoft.Net, what puzzles me is how .Net got there in the first place. It’s not a surprise to hear that the LSE have been criticised on “speed and reliability”, given the environment they were using.
Edited 2010-10-22 00:15 UTC
The LSE has been the butt of a lot of jokes and criticism for quite some time because of their super crappy trading system. It was programmed mostly in C#, uses Microsoft SQL server 2000 (!!) for it’s back end. It was designed with the claims that it would provide trades in under 10ms, a metric it never even came close to. Everybody hated it because it was dog-slow.
Almost 2 years ago it crashed and was down for 6.5 hours, which pretty much spelled it’s doom.
A couple of years ago, when that system went in microsoft were running ads all over the place crowing how the london stock exchange chose windows over linux. Those ads promptly disappeared about the time when their system crashed and cost nearly a day’s worth of trading.
I’m glad to see that Linux is so capable, putting Windows-based servers to shame.
That said, these micro-second trading schemes are all about committing massive financial fraud. It’s known as “high-frequency trading” and Wikipedia has a page:
http://en.wikipedia.org/wiki/High-frequency_trading
It’s a clever way of manipulating stock prices, though a little complicated to explain. Google it for more info. It’s become a gold mine for those who run the servers – they are illegally manipulating stock prices and making millions. This is happening big time in both London and New York.
The law should ban high-frequency trading. All trades should go into the queue and wait a full minute before being executed. That would put an end to this.
Financial fraud my ass. “High Frequency Trading is accounting for over 70% of equity trades taking place in the US”
If it would be a fraud it wouldn’t be so popular but stock exchange is just like gambling it’s for losers. The winners are the fatcats whose doing it since 10 years.
Day trading may be more akin to gambling (it is almost a zero-sum game, after all), but legitimate investing on the stock market is not. Historically, the market has increased in value by an average of 10% year-on-year. If it’s gambling, it’d be like playing a game where E[X]=X*1.1 – good odds, by any measure. And an educated investor can beat the market without too much difficulty. Of course there is risk; there wouldn’t be profit otherwise.
Sorry, but you’re coming across as a certified crazy. You know, like one of these: http://www.youtube.com/watch?v=_c6HsiixFS8
The wikipedia article seems to indicate that there are more positive aspects to HFT than negative ones. Unless you are referring to flash trading? But if so, as the article notes, most major exchanges have banned flash trading, so again I fail to see the problem.
Also, your proposed solution is not the same as the one suggested by the article. The wikipedia article indicates that the proposed regulation would not force the orders to wait until being executed, but would instead prevent the order from being cancelled within the minimum time period. Not delayed, but in force for a minimum amount of time.
I see that my post didn’t go over well. But I stand behind it. Some of you commenters should have done a little googling before replying. But OK, my mistake for not providing more links (see below).
By the way, if any of you think that I made this comment because I’m “anti-Linux”, you’ve got it totally wrong. I use Linux, and love it. The problem isn’t Linux, it’s the crooks on Wall Street. The fact that the crooks use Linux (because it’s a superior OS) doesn’t change the fact that they’re crooks.
Yes, HFT does have some positive aspects, as one of the posters above pointed out. But the stock markets (particularly in New York, I don’t know about London) are rampant with corruption. Anyone who hasn’t heard the term “bankster” by now hasn’t been paying attention. These people don’t lose any sleep ripping off their own clients.
As above poster also noted, I did propose a solution (putting all trades into a one-minute queue, with whopping big fines for non-compliance). I’m not the first one to think of that, but needless to say, the banksters scoff and say that would be another example of over-regulation. Meanwhile, they’ve got their second passports and they’re filling up their bank accounts in Hong Kong and the Cayman Islands.
So here are a couple more links (along with a few notable quotes):
=====================
COMPUTERIZED FRONT RUNNING:
ANOTHER GOLDMAN-DOMINATED FRAUD
http://www.webofdebt.com/articles/computerized_front_running.php
…called High Frequency Trading (HFT) or “black box trading,†automated program trading uses high-speed computers governed by complex algorithms (instructions to the computer) to analyze data and transact orders in massive quantities at very high speeds. Like the poker player peeking in a mirror to see his opponent’s cards, HFT allows the program trader to peek at major incoming orders and jump in front of them to skim profits off the top. And these large institutional orders are our money — our pension funds, mutual funds, and 401Ks…
===================================
Goldman Sach’s High Frequency Trading Scam
http://www.investingblog.org/archives/520/goldman-sachs-high-freque…
One of the biggest market newsmakers recently has been Goldman Sach’s high frequency trading program. The automated system is incredibly complex, using data obtained from the NASDAQ and NYSE markets to front run virtually every trade that happens on the exchange…
Edited 2010-10-22 08:58 UTC
Sorry, but there just isn’t a scam here. The supposed problem, ‘front running’, requires that an exchange be willing to sell information about pending orders; clearly stated in the Wikipedia article and even admitted on ‘Web of Debt’, this is a practice that most major exchanges do not allow.
Also, you need to be more skeptical of what you read online, or just find better sources. ‘Web of Debt’ in particular is lacking credibility. It quotes anonymous sources and people named ‘Tyler Durden’, uses ad-hominems and describes itself as revealing ‘the shocking truth’ and that is ‘eye-opening’. It’s like the National Enquirer of financial news.
There’s only a couple things made clear:
1. It replaces a Windows system while using Linux
2. It’s faster than the old system, at least while they tested it
The article says very little about implementation details: then again, most readers wouldn’t be remotely qualified to understand that.
This doesn’t really speak either way to Windows inferior performance/reliability or Linux’s superior performance/reliability as it does for the implementation of the two trading systems.
Until they measure while controlling for the system-dependent variables, it means… little. And, of course: chances are the system hasn’t had all the same conditions thrown at it that the old system has during its existence: who is to say with the limited amount of real-world testing that it will be more reliable? That remains to be proven with reality. For all we know, there’s some obscure Linux kernel bug that’ll be hit, that will take it down unexpectedly: I’ve seen it happen in past versions of the kernel with massive databases pounding the system.
I’ll come back to this if someone can post a followup that documents what all the variables are, and if the Linux system uses the same logic as the Windows system, or if it is a completely different implementation/design written by different people, but until then, this is merely a PR fluff piece.
I kind of agree. Not only they replaced Windows for Linux, they also replaced the major components of stock exchange software. I think the latter has more impact on performance.
I have a feeling Linux is not the major contributing factor, i.e. replace it with BSD or Solaris and you’re unlikely to see major performance drop.
Over the years all operating systems have been tuned for maximum efficiency, it’s not the kernel developers, but application developers that write inefficient and buggy code.
I read that Windows have problems achieving low latency. I dont know why, but something to do with the TCP/IP stack or something.
Maybe (not sure about BSD though; don’t scale well), but bring back Windows, and everything grinds to a halt again. LOL
This article has more information:
http://lwn.net/Articles/411064/
(you need to be a subscriber or wait 2 weeks)
Is the LSE still at the Royal Exchange these days? I can see the Royal Exchange form my office window. I might go and unplug their network and “crash” the stock exchange.. LOL..
Mainframes are not suited for this low latency tasks. They are dog slow, and instead good at batch processing where latency is not critical. They have high throughput, but bad latency.
Linux/Solaris on x86 has lower latency and is better suited for these demanding tasks.
Some of those big fast performing exchange systems are written in Java.
How do they guarantee that each trade (being a transaction) is persistent in that time frame? In this type of environment, losing a transaction is a major problem; yet guaranteeing persistence requires that the transaction be written to some form of nonvolatile storage prior to completion. Doing this with 175 usec latency seems possible only with very specialized hardware (e.g., battery-backed RAM on the memory bus), and possibly even a customized kernel.
I’ve done Main-Memory DBMS implementation work, and those are incredibly impressive latency numbers for updates, if each transaction is indeed persistent.