Tony Walker, Head of DMA, Merrill Lynch
In the quest for reduced latency (speed to market), every part of the order value chain needs to be examined for its contribution to the time taken to complete the whole round trip time from order entry to exchange order acknowledgement. This includes client's network connection, broker's internal latency, network connection, exchange configuration of hardware and software, order book processing time, even the use of proximity hosting services.
Over the last three years, the benchmark for order round trip times (order entry to delivery of exchanges order acknowledgement) for the client has been reduced from 850ms to sub-60ms, for one particular European exchange. Broker's contributions to round trip timings have come down from 500+ms to single digit, despite large increases in volume, reflecting broad improvements in performance and consistency.
In isolation, these broker improvements could have been be outweighed by the variance of exchange through put capacity. However over the same period, exchange processing times have also improved; one major European exchange's response times have come down from 450ms to 30ms.
The trend towards reduced latency created by both the broker and the exchanges has lead to their contribution to the total round trip being vastly reduced (see diagram Current Latency contribution - 54ms (without proximity hosting)).
That this has led to is a focus upon other components - increasingly clients connect to their brokers via dedicated fibre connections reducing network Service Level Agreements form single digit millisecond times to a micro-second measure.
This has left the network connection to the exchange as an outstanding item to address and the main contributor to the total round trip; in general the connection for exchanges in Europe are reasonably sized direct fibre (or moving to that model) so the network hops are reduced and therefore latency is minimised.
This only leaves Geography; we are limited by the physics constant of the speed of light (or approximately 1/3 speed of light in a glass medium). As we cannot compress the physical constants - the only option is to address the Geography element.
Utilising a data centre close to a European exchange reduces the total round trip significantly - as much as 30% of current latency measures (depending upon actual distance). See diagram (Current Latency Contribution - 36ms (with proximity hosting)).
So if it is that easy why is everyone not doing it? The challenges faced are:
- Distributed architecture for trading engines including Disaster Recovery facilitation
- Multiple outsourced data centres, or expansion of internal data centre and client access
- Requirements for clients (both internal and external) to locate at/close to the brokers architecture and consume market data locally
- Cost of distribution of trading engines, data centre provision, extra network connections
The process over the years has been t consolidate applications and address the scale and latency issues inherent in building a low latency DMA service - the next steps, the fragmentation of the application to local centres is a significant step change from the previous development paradigm. The benefit is worth the challenge.
This trend of round trip time reduction is set to continue through exchange upgrades and competition with alternative, faster, liquidity pools as well as increased competition between brokers to provide the quickest service for their clients. The speed arms race is set to continue with no perceivable end in sight.