Activ Outlines Plans for FPGA-, AMD-Based Ticker Solution
From A-Team Group’s Market Data Insight, April 2007: Activ Financial appears to be first out of the box with a hardware-based solution to clients’ low latency challenges using Field Programmable Gate Array (FPGA) technology. The company plans to formally introduce its ActivFeed MPU (Market Data Processing Unit) hardware accelerated feed processing platform at an event at the Nasdaq MarketSite in New York on May 14.
The company has been working to port its market data ticker plant solution directly on to silicon using an FPGA co-processor and Applied Micro Devices Inc.’s AMD64 technology featuring Direct Connect Architecture and HyperTransport technology. AMD appears to be making a serious push into the financial low latency area, with a separate collaboration with Activ competitor Wombat Financial Software. The company’s Torrenza initiative aims to couple hardware-accelerated co-processors with CPUs via HyperTransport.
According to Frank Piasecki, president at Activ, the company will offer the hardware-based low-latency delivery solution alongside its existing software-based ticker plant/feed handler offerings. Activ will provide the new capability – in the form of installing a co-processor card into the motherboard slot – as an option or upgrade to their existing infrastructures. Piasecki stresses that such an upgrade will not require modification of the Activ feed handling software. Piasecki believes the move into hardware solutions is a step that offers key latency benefits as well as total cost of ownership advantages that financial institutions will find difficult to ignore. “I can’t imagine lots of people not doing this,” he says.
The HyperTransport technology essentially reduces the latency to bus speeds by making the co-processor a peer to the CPU. Mike Dunne, CTO of Activ, says the new MPU approach could potentially reduce an eight- to 12-server installation into a single hardware-accelerated device.
Piasecki believes Activ’s decision to work with AMD and the HyperTransport consortium allows the company to avoid a stumbling block that he believes has heeded progress by others attempting hardware-based low latency solutions: sole-source component vulnerability. By basing its solution on as close to standard equipment as is available in this area, Piasecki reckons, ActivFeed MPU meets the ‘hardware comfortability’ criterion so important to Tier 1 financial institutions in their enterprise data considerations.
Piasecki believes the broader shift toward hardware-based solutions ultimately will take the input/output consideration out of the low latency solution set. As a result, he suggests, the low latency dilemma will no longer rest with the network or infrastructure element of the overall trading architecture, but will in effect shift downstream, to the order routing and order management system layer. He believes that the presence of major enterprise players like AMD in the space will exert a gravitational pull of sorts on ISVs and clients alike, bringing them into what will emerge as a FPGA community.
Activ emerged from three-and-ahalf years of stealth-mode operation last summer. Founded by former executives of A-T Financial, a Chicago-based data feed handler, and ICV Ltd., a U.K. market data vendor, the company has to date pursued a three-pronged product line-up, all underpinned by a common delivery architecture known as ActivMiddleware.
This architecture is used to support direct exchange feed access – via the ActivDirectFeed offering – a consolidated datafeed known as ActivFeed, and trader workstation display – known as ActivWorkstation. The line-up also includes its Open Content Management Platform, which allows Activ to incorporate value added services like client-contributed data and calculations, last-trade information, time-series data and news archiving into its services. The addition of ActivFeed MPU also builds on the company’s stated aim of becoming a full-service provider in the low latency data distribution space. As well as providing stripped-down direct exchange access – using both software- and hardware-based solutions – Activ can also offer the value-added functions – vendor of record, exchange permissioning and reporting, and so on – that the market expects.
Tell us what you think - leave a comment below: Is your firm going to be comfortable with a hardware-based ticker plant? Or does FPGA/co-processor technology represent a risk? What is the trade-off? Would you want Intel to respond with a similar co-processor approach?
May 9th, 2007 at 2:24 pm
In 2003 a company was founded as a spin-off from Washington University to develop some of these new computing techniques. The company in question was Exegy. (There was some movement of talent between Reuters St Louis and Exegy in late 2005 including Scott Parsons, now the Chief Scientist at Exegy). By the summer of 2006 Exegy had introduced hardware acceleration to the world of market data (SIA show demonstration). I joined the company myself in November last year.
Hardware acceleration is revolutionary in that it overturns conventional IT solutions to solving market data problems. Because it is a big change there will be some resistance and legitimate questions about the technology. However, the economics are compelling. For example, the old approach to ever escalating market date rates is to scale horizontally by buying more commodity computers. But the continuing growth of market data makes this an increasingly expensive proposition. Well designed hardware acceleration ticker plant technology can massively increase message per second headroom and reduce latency and allow the infusion of calculated values to the data stream output. So you can have your cake and eat it and get a cherry on top.
The early adopters are in the trading community. Traders and competing exchanges demand the fastest possible delivery, analysis and execution of market data. It will be a brave financial institution that chooses to ignore the innovation.
This revolution is similar to the introduction of new materials technology in the 1970s to world of tennis. Once well designed metal and graphite models were introduced in the 1970s no professional player could afford to be under-equipped with old style wood racquets. The new material racquets were expensive in the early days but the power improvement and the much larger sweet spot made them indispensable. Anyone remember the Dunlop Maxply? The last one was produced in 1983.