Sybase Business Intelligence Solutions - Database Management, Data Warehousing Software, Mobile Enterprise Applications and Messaging
Sybase Brand Color Bar

There’s plenty of room at the bottom

Thought I’d write a quick missive after trawling through the latest info on low latency trading. Apologies to Richard Fenyman for the borrowed headline (he did a seminal speach on nanotechnology). The area I am referring to is low latency trading. The current arms race to the bottom of latency arbitrage is nearing a new phase. The quickest players are using FPGAs (Field Programmable Gata Arrays) for Microsecond latency and I suspect embedding them directly into switches in the exchange is going to be about as quick as that can get. In fact just build a network of routers with trading FPGAS in the exchange data centre and that’s about as quick as you can get without getting more exotic into the realms of optical computing etc . Of course in totality you are looking at a number of different processes end to end , and reducing these to get end-to-end latency down. Anything using normal PC Based architecture (network cards included) is nowhere near this in practical terms. High microsecond are achievable using off the shelf operating systems and hardware.

To recap.
1 millisecond (ms)
= one thousandth of a second (200 milliseconds for the blink of an eye, 300 milliseconds to react to something)
1 microsecond (us) = one millionth of a second
1 nanosecond (ns) = one billionth of second (one machine cycle of 1ghz processor.)

Millisecond or high microsecond is a reasonable goal for CPU/OS based processes. Anything less and really you are constrained by the architecture of the server itself (i.e. the number of hops from network to OS to CPU/RAM and back again, including traversing the TCP/IP stack @ 10 milliseconds ). Anyone claiming Pico or Nano on and normal PC architecture is just not telling the story. And when you move to specialised architectures you lose flexibility. The number of people who can design and build those systems reduces dramatically.

Why does this matter? Whilst in the case of high frequency trading or basically ultra high speed-based arbitrage this is incredibly important. It is becoming a race that takes exponentially more money and ultimately the physics game can be played by anyone, but only with the deepest pockets, and fractional advantages are eroded in the next cycle of development . (Yes , were happy to embed our CEP anda analytics in a co-lo to gain as much speed as possible for apps that need it) , but I am talking about the other 95% + of applications that are not ultra low latency.

For the huge number of use cases and applications outside of this zero latency arms race their is a significant amount of value to be exploited in making the rest of the Trading system run quicker. There are large numbers of applications doing end of day calcs instead of intra day for example. Running in 10 minutes or even 5 seconds what might take 4 hours is potentially more ground breaking that reducing something from 10 to 9 microseconds. Index calculations that are currently done at close of exchange. Moving to an intra day model would add huge value for some of them.

Market data systems are currently being redesigned for increasing flows, adding CEP to them to intelligently process, filter and cleansed data in milliseconds. Then caching in local in-memory databases (al la Sybase IMDB!) and refreshing from CEP for low latency market, that’s game changing. Pre , post , order execution and risk systems can all benefit from doing things orders of magnitude quicker. Change hours to minutes, minutes to seconds and seconds to milliseconds. Sybase RAP can address these by using specific engines for each role (low latency analytics and streaming, in memory database and high speed columnar analytic stores in a single platform using off the shelf hardware . And using tools like the CEP and RAP developers can quickly churn out high performance low latency code using rapid development tools that can provide your market edge, reducing the time to turn around new models and strategys and providing IP differentiation.

And there are many capital markets systems that can benefit from this architecture. There’s plenty of room at the bottom!

Share and Enjoy:
  • Digg
  • Sphinn
  • Facebook
  • Mixx
  • Google Bookmarks

Leave a Reply

Sybase privacy policy