2
www.securitiesindustry.com What Makes Markets September 18, 2006 BY JEFFREY KUTLER H igh-performance database devel- oper Kx Systems has completed a comprehensive installation at Cantor Fitzgerald, which in turn is providing a rare glimpse into the decisions that are accelerating the race to analyze and trade on market data in milliseconds—and into why a major banking and trading firm chooses one such system over another. Palo Alto, Calif.-based Kx plans to announce its Cantor deployment as early as today, though the New York securities firm, which has a long-standing reputa- tion as a technology innovator, has been working with the Kx kdb+ database soft- ware since November. Many such instal- lations are never publicly announced or elaborated upon because the technology confers a competitive edge and firms do not want to tip off the competition. But it is an open secret that many big investment banks, asset managers and hedge funds use Kx in sophisticated, high-speed analytics. Founded in 1993 by Arthur Whitney and Janet Lustgarten, respectively the CEO and president, the company specializes in systems that cap- ture and analyze data at speeds and vol- umes that traditional, relational database management systems and query languages could not match. Cantor traders are using the kdb+ data- base, and the proprietary Q programming language that goes with it, to track portfo- lio risk and trades in real time. That kind of immediacy has become a much-sought- after goal throughout the trade life cycle. Cantor sees it as a boon to gauging price moves and being able to move in and out of the market quickly when implementing trading strategies. “Everyone talks about real-time risk management, but with kdb+ we actually attain it,” Cantor SVP Jacob Loveless said in a press-release statement issued by Kx. “We’ve gone beyond simply knowing the risk to being able to act on it in the moment.” Cantor initially wrote a back-testing program with the software in two days, then built other historical and real-time data capabilities on top of the architecture, which because of its integrated nature assures no time lag— zero latency—between data capture and analysis. “The ability to integrate real- time publishing with historical searches made kdb+ the obvious choice for our new analysis system,” said Loveless. Crowded Field Kx is hardly alone in trying to aid Wall Street’s drive toward low-latency data and trading infrastructures. Its niche is occu- pied by the likes of Vhayu Technologies, whose Velocity system is geared for tick data processing and algorithmic trading and has been making inroads internation- ally, and a new wave of ultra-high-volume event and stream processing systems from Aleri Labs, GemStone Systems, StreamBase Systems, Progress Software Corp. and others. The traditional rela- tional database leaders—notably IBM Corp., Oracle Corp. and Sybase—have taken notice and are touting their staying power and research-and-development prowess. Oracle in June 2005 acquired TimesTen, a Kx and Vhayu rival. Kx chief technology officer Simon Garland said in an interview last week that the crowding of the field is good news for the marketplace, not only because the competition spurs perform- ance advances, but also in bringing credi- bility and notice to what had been a rela- tive technological backwater. “Five years ago we were a lone voice,” said Garland, who joined Kx in 2002 after working in the search engine business and in Credit Suisse’s risk management area. Kx was a pioneer in envisioning what is now kdb+, with capacity for “many, many terabytes” (trillions of characters), and “we’ve been at this long enough to establish credibili- ty,” said Garland. The most recent kdb+ release, version 2.3, was announced in March. Its multi- thread capability allows firms to take advantage of the latest generation of par- allel processing using multicore, multi- CPU (central processing unit) chips from the leading semiconductor manufacturers. “Clients are constantly investing in new hardware and applications to keep up with increasing data volumes,” CEO Whitney, a veteran of UBS and Morgan Stanley, said in the 2.3 announcement. “It’s not unusual for a program trader to crunch through gigabytes [billions of characters] of trade and quote data in seconds. Kdb+ 2.3 gives them the capacity to process large amounts of streaming, historical and real-time data simultaneously.” Kx is now gearing up for another leap in performance with quad-core chips. The next release is likely by year-end, said Garland, adding, “programmers will be excited by more multithreading.” Tools Are Critical “If the infrastructure is the engine for market data, the in-memory and time- Kx Database at Cantor Makes Real-Time Real

Sec Indus News

  • Upload
    jd38

  • View
    6

  • Download
    1

Embed Size (px)

DESCRIPTION

Security Industry News

Citation preview

  • www.securitiesindustry.comWhat Makes MarketsSeptember 18, 2006

    BY JEFFREY KUTLER

    H igh-performance database devel-oper Kx Systems has completed acomprehensive installation at CantorFitzgerald, which in turn is providing arare glimpse into the decisions that areaccelerating the race to analyze and tradeon market data in millisecondsand intowhy a major banking and trading firmchooses one such system over another.

    Palo Alto, Calif.-based Kx plans toannounce its Cantor deployment as earlyas today, though the New York securitiesfirm, which has a long-standing reputa-tion as a technology innovator, has beenworking with the Kx kdb+ database soft-ware since November. Many such instal-lations are never publicly announced orelaborated upon because the technologyconfers a competitive edge and firms donot want to tip off the competition.

    But it is an open secret that many biginvestment banks, asset managers andhedge funds use Kx in sophisticated,high-speed analytics. Founded in 1993 byArthur Whitney and Janet Lustgarten,respectively the CEO and president, thecompany specializes in systems that cap-ture and analyze data at speeds and vol-umes that traditional, relational databasemanagement systems and query languagescould not match.

    Cantor traders are using the kdb+ data-base, and the proprietary Q programminglanguage that goes with it, to track portfo-lio risk and trades in real time. That kindof immediacy has become a much-sought-after goal throughout the trade life cycle.Cantor sees it as a boon to gauging pricemoves and being able to move in and outof the market quickly when implementing

    trading strategies.Everyone talks about real-time risk

    management, but with kdb+ we actuallyattain it, Cantor SVP Jacob Lovelesssaid in a press-release statement issuedby Kx. Weve gone beyond simplyknowing the risk to being able to act onit in the moment. Cantor initially wrotea back-testing program with the softwarein two days, then built other historicaland real-time data capabilities on top ofthe architecture, which because of itsintegrated nature assures no time lagzero latencybetween data capture andanalysis. The ability to integrate real-time publishing with historical searchesmade kdb+ the obvious choice for ournew analysis system, said Loveless.

    Crowded FieldKx is hardly alone in trying to aid WallStreets drive toward low-latency data andtrading infrastructures. Its niche is occu-pied by the likes of Vhayu Technologies,whose Velocity system is geared for tickdata processing and algorithmic tradingand has been making inroads internation-ally, and a new wave of ultra-high-volumeevent and stream processing systemsfrom Aleri Labs, GemStone Systems,StreamBase Systems, Progress SoftwareCorp. and others. The traditional rela-tional database leadersnotably IBMCorp., Oracle Corp. and Sybasehavetaken notice and are touting their stayingpower and research-and-developmentprowess. Oracle in June 2005 acquiredTimesTen, a Kx and Vhayu rival.

    Kx chief technology officer SimonGarland said in an interview last weekthat the crowding of the field is good

    news for the marketplace, not onlybecause the competition spurs perform-ance advances, but also in bringing credi-bility and notice to what had been a rela-tive technological backwater. Five yearsago we were a lone voice, said Garland,who joined Kx in 2002 after working inthe search engine business and in CreditSuisses risk management area. Kx was apioneer in envisioning what is now kdb+,with capacity for many, many terabytes(trillions of characters), and weve beenat this long enough to establish credibili-ty, said Garland.

    The most recent kdb+ release, version2.3, was announced in March. Its multi-thread capability allows firms to takeadvantage of the latest generation of par-allel processing using multicore, multi-CPU (central processing unit) chips fromthe leading semiconductor manufacturers.Clients are constantly investing in newhardware and applications to keep up withincreasing data volumes, CEO Whitney,a veteran of UBS and Morgan Stanley,said in the 2.3 announcement. Its notunusual for a program trader to crunchthrough gigabytes [billions of characters]of trade and quote data in seconds. Kdb+2.3 gives them the capacity to processlarge amounts of streaming, historical andreal-time data simultaneously.

    Kx is now gearing up for another leapin performance with quad-core chips.The next release is likely by year-end,said Garland, adding, programmers willbe excited by more multithreading.

    Tools Are CriticalIf the infrastructure is the engine formarket data, the in-memory and time-

    Kx Database at Cantor Makes Real-Time Real

  • 2006 SourceMedia, Inc. and Securities Industry News. All rights reserved. SourceMedia, One State Street Plaza, New York, N.Y. 10004 (800) 367-3989

    series databases are the turbochargers,Tom Price, a senior analyst at Needham,Mass. research firm TowerGroup, wroteof the Kx-type technology in a May2006 report on the market data land-scape. Low-latency data receipt anddistribution are critical, but they meannothing without tools for analyzing thedata and translating it into actionablemarket opportunity.

    Price cited Bear Stearns & Co.s selec-tion of Vhayu early this year as indicativeof how major firms are gearing up formillisecond-level response require-mentsand of the effectiveness and scal-ability of Vhayus Velocity software. ButBear Stearns was sparing in its publiccomments about its choice.

    Privately held Cantor Fitzgerald,known particularly as a leading institu-tional government securities brokerageand, technologically, for its spin-off ofthe publicly listed eSpeed trading sys-tems subsidiary, is openly praising Kxfor its platforms speed, flexibility andefficiency, as well as the support andrapid response time of Kxs sales and

    services partner, U.K.-based FirstDerivatives. Garland said that Cantor istaking advantage of the ability toaccommodate tick data, historicalqueries, multi-asset-class trading andother needs on a single platform, solv-ing the entire puzzle at once.

    Emphasizing FlexibilityPressed, in an exchange of e-mails,about why Cantor went with Kx,Loveless replied: We looked at a hand-ful of vendors and in-house technolo-gies before choosing Kx. We chose Kxfor one simple reasonwe couldnt findanything that was faster and more flexi-ble. We are provided access to dailybuilds of new versions and have directinput into features and performancerequirements going forward. That abil-ity to customize a product as our needschange has immense value.

    The focus on low latency and the abilityto respond in fractions of a second is lessdirectly about profit-and-loss than aboutmanaging overall market risk, saidLoveless. Markets gap, and gap quickly. If

    your mark-to-market is ten seconds old, itsjust not true. As exchanges and ECNs [elec-tronic communications networks] becomefaster, lower latency will become more andmore important. This industry is in a tech-nology arms race with regards to latency.

    He said that as with any new product,there are scalability and disaster-recov-ery concerns. Kxs solution to both wasa simple, high-speed integrated messag-ing system, he said, adding, If we didhit any limits, I would expect Kx to riseto the challenge. He recalled dis-cussing with Kx a complicated analyticthat took about 20 minutes to calculate,but the Kx system shortened that to 20seconds. Its hard to hit scalability lim-its when you can increase performance3600 times by working with your ven-dor, said Loveless.

    Integrating a proprietary-languagesystemLoveless characterized the logicas 95 percent in Q, 5 percent in Cwithlegacy systems was no small task, saidLoveless, but it came easily enough. Wefound that kdb+ forces one to simplifyintegration points. n