Reuters Breaks One-Millisecond Barrier with RMDS. Achieves 170,000 updates per second through a single CPU

Reuters, the global information company, today announced that in a controlled test environment, the core of the Reuters Market Data System (RMDS) sustained speeds that reduced latency to less than one millisecond. In addition, RMDS demonstrated an ability to supply data to a single application at over 170,000 updates per second.

RMDS manages financial content in real-time across the client’s enterprise. The system is currently deployed or in the process of rollout at more than 270 sites worldwide.

Mike Alsip, Head of Product Management for RMDS at Reuters, said: "Our latency profiling shows that RMDS provides a very fast way to distribute data to latency-sensitive applications in areas such as automated trading. And our capacity tests show that it can exceed the throughput requirements of these demanding applications well into the future. Importantly, RMDS achieves this without compromising its other hallmark features, such as reliability, scalability, and a permissioning mechanism approved by nearly all of the world’s exchanges."

Octavio Marenzi, Founder & CEO of analyst firm Celent Communications said: "Reuters focus on latency and other requirements for equities electronic trading is important within the industry. This shows that they are committed to the evolving needs of their customers, and are proactively looking for ways to address the significant growth in market data rates."

Latency
Laboratory tests revealed that the latency between the entry of data into RMDS and arrival within an application was less than one millisecond, under the traffic loads that are typical of US equity and options markets. Data latency is important because an increasing number of computer-driven trading strategies require the ability to act nearly instantaneously on fleeting price movements.

Larry Tabb, Founder & CEO of the Tabb Group said: "Advanced equity trading technology is radically changing the market data business, forcing data providers to focus on extremely low market data latency at extremely high volumes."

RMDS was configured in a “flat” architecture, with just one network "hop" between the data source and consuming application. The flat architecture enables applications to run directly on an RMDS backbone node, as opposed to the more traditional tiered configuration.

Throughput
The tests also showed that RMDS could route data from the backbone into a single consuming machine at 170,000 updates per second, while using just a single CPU. Because RMDS is a distributed system capable of spanning multiple multicast backbones, it has no theoretical limit on aggregate throughput or the number of applications it can support. However, throughput to a single application is a key metric in markets such as equities and derivatives, where applications often need to consume large amounts of data.

Client environments
The test results announced today are the first results from an optimization and tuning program for the core RMDS components. In these tests, the RMDS servers were deployed on 3.06 GHz Intel Xeon platforms running Linux, communicating with each other across a one gigabit-per-second Ethernet switched network. The RMDS software consisted of versions that are available to clients today.

The laboratory tests simulated a protected, low-latency RMDS hub that a client can implement for high-speed applications. The flexibility of RMDS means that this hub can also feed a client’s existing trading floor configuration without requiring an architectural overhaul.

Latency and throughput are subject to a number of factors specific to each client environment. Clients are encouraged to contact Reuters for assistance in optimizing their production environments using RMDS.

Become a bobsguide member to access the following

1. Unrestricted access to bobsguide
2. Send a proposal request
3. Insights delivered daily to your inbox
4. Career development