Building the Solution: Legacy to Contemporary

Posted on June 5th, 2008

Working with the Thomson Reuters RDMS in the Intel fasterLAB is akin to adjusting the spoiler on a Formula 1 race car by one degree to improve performance, says Paul Gow of the consultancy, CJC Ltd.

“The RMDS is very fast and it is a very resilient piece of software. It has a lot of parameters that can be tweaked to enhance the system, and a lot of those parameters are hidden and people don’t know they exist.”

At the Intel fasterLab, CGC works with a base system of RMDS on which firms can use to test their desktop applications.

“So a client will come in with an application that they want to go faster than the data coming out of the RMDS pipe. We will tune the system and make it go faster and hopefully make it break their application with the update rates and then we help them move forward.”

One of the most common sources of latency in trading is in-house applications that can’t keep pace with market data. Firms write their APIs to the Reuters data, but if the API in the application can’t handle the data refresh rates, the machine can come to a grinding halt.

As a veteran in the field, Gow is leery of latency claims from vendors and how relevant they are to a production environment.

“Anyone can put one box at the end of a piece of wire and push data down it and say we have zero latency between those two boxes. But when you put it into client network with switches and firewalls and hops, are those useful measures?”

As market data experts, CJC examines client systems, designs, and builds and offers support, training, and administrative services.

To improve latency, the firm would look at the overall design of a system, its configuration, the amount of network connectivity, network speed, and performance of the computers. At the Intel fasterLab it will have access to equipment from, for example, Sun and Cisco, choices of other hardware and infrastructure switches - and be able to replicate client production systems.

“They can connect their application and start tuning to system to make it go faster.”

Gow also expects to work with virtualisation, leveraging Intel’s many close partners in this area such as VMware, to see if it can meet the demands of high-speed trading with the benefit of virtualisation and not lose performance. Although some financial services technologists think virtualisation will be too slow for trading, Gow wants to test it and look at results. Early results in the fasterLAB from Cohesive using AMQP sw and Elastic Server showed effective distribution of trading scale workloads (using Pantor’s OPRA simulated feeds) – and subsequent performance runs with other AMQP implementations from RedHat and iMatx have demonstrated that this open source software is potentially ready for prime time in the testing front office environment.

“New technology arrives every day. Until you can benchmark it against the system then we don’t know how fast it can be. You hear stories that virtualisation slows down processing, but does it? With the Sun and Intel guys at the lab – they are serious technical people – so we will see if we can overcome these problems - we can see significant benefits from access to today’s leading technologies via the fasterLAB - we can contrast and compare.”