To print: Click here or Select File and then Print from your browser's menu

This story was printed from silicon.com, located at http://www.silicon.com/

Story URL: http://www.silicon.com/retailandleisure/0,3800011842,39169164,00.htm

Betfair betting engine goes supersonic
Case study: Extreme transaction processing at nose-bleed speeds...

By Julian Goldsmith

Published: Thursday 15 November 2007

Betfair has developed a new core betting engine that can process staggering volumes of bets per second and significantly reduce the cost of each transaction.

Agenda Setters 2007 - Top 20

Find out who made it into silicon.com's Agenda Setters top 20 by clicking the links below…

1. Mark Zuckerberg
2. Steve Jobs
3. Eric Schmidt
4. John Chambers
5. Ashley Highfield
6. Nicholas Negroponte
7. Niklas Zennström
8. Diane Greene
9. Jonathan Ive
10.Viviane Reding
11.Paul Coby
12.Marc Benioff
13.Emily Bell
14.Larry Ellison
15.Jeff Bezos
16.Ben Verwaayen
17.Nandan Nilekani
18.David Yu
19.Satoru Iwata
20.Mark Hurd

The technology the company came up with is so impressive, it was the clear winner of the CNET Networks UK Business and Technology Retail & Leisure Technology Project of the Year Award for 2007.

Betfair is an online betting exchange - a peer-to-peer gambling site acting as a broker between parties for the placement of bets. The concept is similar to a stock exchange.

Matt Carter, Betfair director of architecture and advanced technology, explained to silicon.com customer trends are moving towards a high volume of small-value bets, which, in turn, produce small-value commissions. He said the only solution was to develop a system that could process exponentially more transactions for a lower cost than traditional betting engines.

Carter said: "It wasn't a case of whether we needed to do this, we have to do it. It's a question of survivability."

Betfair CEO David Yu set the company's R&D team the task of increasing throughput by 100 times for free. Project 100X took two years to develop and was run both internally and with three other partners, in what Carter calls a "bake out" - whichever team came up with the best prototype would get the investment for the rollout.

The budget for the entire project was less than £1m over two years.

In the end, the R&D team came up with a betting engine, called Flywheel, that could demonstrate a throughput of almost 100,000 transactions per second, while also reducing the cost per transaction by 200 times.

Even with a traditional betting engine, Betfair processes five million transactions per day - much more than the London Stock Exchange's transaction processing system is capable of.

The R&D team expects one million trades per second to be possible through Flywheel, which it estimates is the equivalent of the entire combined annual global equity trading volume being processed in a matter of hours.

However, for Carter, the key achievement is the cost levels. He explained the whole system runs on two servers with an approximate cost of £25,000. This, he said, is in comparison to a high street bank with a similar throughput load that will typically use a mainframe costing many millions of pounds.

The engine hasn't been installed into Betfair's core systems yet and Carter couldn't give any precise deadline for migration. In the interim, the company is about to implement Flywheel Lite, which will unleash some of the performance of the new system, while sweating the assets of the old betting engine. Carter doesn't see the need to make use of all that transaction processing power just yet. It is at his disposal when the company needs it.

The modular nature of the architecture, which Carter compares to a service-oriented architecture, and the transaction processing power, gives Betfair the freedom to innovate and come up with new products to offer punters.

Carter said: "If we hadn't found a way of increasing transaction volumes, while lowering the cost per transaction, Betfair would have been forced to change its business model."

Betfair has applied for a patent for Flywheel and is in discussions with other organisations, both in the private sector and education, about how they can share knowledge on extreme transaction processing.