Remember me

Register  |   Lost password?

The Trading Mesh

FPGA on the Cloud for the Trading Industry

Tue, 12 Aug 2014 09:53:57 GMT           

Introducing the New Normal

By Nicolas Karonis, Managing Director Business Development, Enyx


The Cloud has always been associated with flexibility, customized service on demand, no upfront costs, ease of upscaling, no IT footprint and very competitive running costs thanks to the sharing of resources.

FPGA in the trading industry has always been associated with unbeatable latency and zero jitter but also very specific and hard to find skillsets, high initial investment costs and high ongoing costs, for maintaining, upgrading or upscaling a platform.

Also - and this goes beyond FPGA - there has been a wide perception in the trading community that latency issues have driven a technology war, a “race to zero” that distorts the market and provides a systematic and unfair advantage to those happy few who can invest tens of millions of dollars in both hardware and infrastructure.

However, FPGA technologies have now evolved to the point that they have become mature enough to be offered on the Cloud, essentially bringing to the trading community the best of both worlds and offering a level playing field to all traders.

Let’s start with Market Data, where low latency with no jitter has become the new normal. Even for strategies that are not hyper sensitive to extreme low latency, nobody wants to be always 3rd in the queue or not to be able to trust the data he gets when there is a market event resulting in a volume burst.

FPGA solves the latency and jitter issue, the Cloud democratizes those services.

More specifically, with the combination of managed FPGA infrastructure and services and hyper fast connections between Data Centers (also using wireless connections when appropriate) the Cloud allows you to tailor the way you receive FPGA accelerated feeds. 

But first you have to address the data content issue. 

You can tailor the feeds your automated algo engine will receive (through 10Gbs UDP multicast thus re-using your existing NICs) by configuring two parameters for each one of your algos: Trading Universe (the symbols that you trade or that are relevant to your trading decisions) and Market Depth. The trading algo will be able to dynamically adjust those settings through a very lean API written in a language familiar to the trading community such as C++.

Second you have to address the algo topology issue.

 The algo engine will usually be located in the exchange colocation facility where the orders are sent, but the data influencing the order trigger decision will be coming from multiple sources and locations. By combining FPGA and millimeter-wave technologies you can package best-in-class latencies in both data processing and inter-site connectivity. With proper filtering and the best binary formats, you can receive on a 10Gbs standard Ethernet port the exact data that you need wherever you need it without worrying about latencies or being at a disadvantage compared to the big guns.

What’s coming next?

The FPGA trading engine on the cloud. This includes pre-trade checks and essentially allows you to take care of order management and risk with the latency cost of a bump on the wire (less than 1.5 µs).

More about that on our next blog.


, , , , , , , , , , , , , , , , ,