Late last year, Intel held the first in a series of fasterFS events in London that explored the direction and opportunities created by quad core technology. Read how senior financial markets executives from Atos Euronext, Reuters, Credit Suisse Lehman and the London Stock Exchange plan to re-engineer their trading application infrastructure while significantly reducing the server and environmental footprint of their current operations.
Intel FasterCITY Event 27 Nov’06
Intel fasterFS panel
- Nigel Woodward, Financial Services Director UK, Intel (moderator)
- Ray Mulligan, Global Head of FX IT, Credit Suisse
- Robin Paine, CTO, London Stock Exchange
- John Oddie, Global Head Exchange Business Unit, AEMS (Atos Euronext Market Solutions)
- Kevin Covington, EVP Product & Business Development, BT Radianz
- PJ Di Giammarino, CEO, JWG-IT
- Peter Moss, Global Head of Enterprise Solutions, Reuters
- James Watson, Head of Direct Access Trading Solutions, Lehman Brothers
Nigel Woodward: With all the various pressures – both competitive and regulatory – will the quest for speed ever stop? Where can extra speed be found?
James Watson: One of the most important things is latency and the complete round trip speed, especially in the Stat Arb business. Once you get into the algorithms there is a cross over into computational needs rather than round trip delays. We’re at 130-135 milliseconds and getting to 125 will be another dimensional step. We’re interested in making the algorithms even deeper which requires more computational ability – it’s a direct ratio between analytical capability and processing power, and therefore speed of automated execution.
Ray Mulligan: Latency depends on how you define it. I could deal with 10 second latency as long as Lehman’s manage 15. It is relative speed or competitive advantage. The search for speed won’t stop until it stops yielding an improvement in the effectiveness of algorithms to detect and exploit arbitrage opportunities.
Robin Paine: As an exchange, absolute latency is important but consistency of latency is more valuable particularly when new order types are introduced. We are running a market – hence predictability, consistency and reliability are critical features.
NW: And this will lead to competition between the execution venues?
John Oddie: We will see in Europe what we saw in the US - electronic venues competing on performance and price. MiFID is an exercise to create competition and you can only compete on price and performance. If you offer price and performance, you will attract liquidity in a self supporting mutuality. For example, in the US, the smart routing technology became very intelligent and monitored how fast the ECNs were performing. The result was that traders began dropping ECNs and moving to another venue if the performance dropped off.
NW: But with so many venues will the levels of performance ever match demand?
Peter Moss: We deal with 240 exchanges in equities alone on a daily basis and that number is likely to increase with the arrival of MiFID. We see a continuing demand for venues to be aggregated in this way on a global basis. But a consolidated network can not deliver the ultra low latency required by many automated trading applications today and so it has to be supplemented by feeds direct from the venues to the customer site. With this combination and the right choice of low latency technology it is possible to match the demand for performance.
NW: Is there an operational risk in stretching the envelope a little too fast?
JO: The challenge is how to benefit from what is happening if you have over architected solutions that cannot take advantage of multi core technology? You have to think about software tiers of threading, N+1 type technology and small units of relative discrete processing. The message I am seeing from the vendors is virtualisation and threading as the latest offering to provide more horsepower.
NW: How are the vendors reacting to market demands? The supply side of the market is talking about ecosystems of vendors – what does this mean ?
Kevin Covington: We’re seeing growth and consolidation of venues where liquidity exists. The constant is growth in volumes of data. The key thing is managing the end to end trading process which will take place across the systems of a number of vendors – the ecosystem. To take advantage of Intel quad core technology and other innovation people are going to have to participate with this range of vendors to tune the complete environment. Many of these are quite small so new partnering skills will have to emerge. However the low latency prize is so large, cooperation is given an adrenalin shot.
NW: And what is happening in the network, connectivity and hosting space right now?
KC: It’s an arms race, and as a network and hosting provider we are in it too - with sell side firms offering levels of service to their clients that previously we had never encountered. But today’s venues may not be tomorrow’s. People are searching all the time to find the next piece of the process that they can improve, in order to provide their customers with a better service. Our networks have to be faster, and our service levels have to increase to offer the chance of higher application performance.
NW: This conversation has touched on MiFID. Are we in a state of flux?
PJ: MiFID will go on for quite some time and there will be winners and losers. This discussion is as much about how you run the bank – firms can’t spend 25% of their IT budget on change projects. MiFID is a complex project, shifting to new technology without a lot of testing time available. The trade execution venues and data vendors in an open competitive market will have a lot to play for and will be willing to devote a serious amount of resource to this.
NW: With the number of venues increase significantly is there a massive data management explosion in the making?
RM: Intel can make processors that are 3, or 4 times faster, but you need to take into account that software engineers can easily make things 100 times slower. So you really have to start with the software engineering – so here I am endorsing the ecosystem mix mentioned earlier, you have to look at all the variables. The skill however lies in identifying which area will make the highest contribution and applying technology resource and investment accordingly.
NW: But can DMA be made any faster? What are clients demanding?
JW: The clients tell you how to do it and you hang on for dear life! They call the shots. The DMA firms face this challenge from clients every day. Clients are not interested in the technology arguments being discussed here. They just want to remain innovative and have to keep getting faster. The hedge fund firms have hired the best engineers, smartest software people and every time we deal with a client it is a different scenario each time. So they challenge us which means we have to challenge our suppliers.
NW: How is Reuters responding?
PM: Reuters has had to innovate around our direct feeds and our market data systems to get the performance our clients are demanding. We are working directly with the Intel engineers to improve software performance and we are finding that this is what it takes to meet the performance our clients demand. This is a change – to be frank, a few years ago Intel just did not have the technology or the attitude, however today the new chips and focus on trading impress us and are capable of making a big difference.
NW: Kevin, are you looking at engineering a faster infrastructure to address some of the software issues raised?
KC: The first phase was to tune our network performance, build SLAs that ensure that you can predict latency and monitor performance. Now we’re seeing interest in how you can improve the performance of the applications themselves, either bank built applications or third party applications such as exchange gateways. This optimisation with hardware acceleration companies is very much early days – being done on a point-by-point basis. We need to think about distributing the applications in a new way, which would encourage firms to get closer to the processor suppliers such as Intel to look at the entire process – at depths not previously considered.
NW: Old systems are now being asked to process quantum leaps in volume. What is the policy in terms of how you retire those applications? It implies a fundamental change in process.
RM: We constantly invest in keeping infrastructure up to date. When a buy side client gives us an order we run the algo engine as close as possible to the execution venue, this will be very important when you might have a matching engine at a different location. A lot of work happening over the next 18 months is to make sure that we can take advantage of the proximity services being offered.
NW: John you are running a very broad business. Where do you see change?
JO: We haven’t talked very much about cost. It’s not just about performance but about price performance. When an exchange executes, what they may actually have is ten or 100 times as many order messages as the volume of trades they execute. The volume traded only goes up by factors of two or three orders, well behind the rate of order growth.
So, the question is where do you go with power? You hear stories of Google building a server farm with a million units and growing at a rate of 1000 a day! A lot of players in the City will have real issues with managing the heat and power necessary to meet the operational performance required by clients. This will be a challenge for the vendors, Intel and the exchanges.
The data implications of MiFID and trading – and the technology requirements around those — are phenomenal. And this is going to drive new data centre and data management approaches.
NW: So what does 2007 – and beyond – have in store?
RM: The rise of e-trading and algo trading in the FX space.
RP: The challenges for LSE will be to continue innovate on our new technology platforms providing consistency and predictability in terms of latency and availability.
JO: An ‘arms race’ in Europe between the major exchanges and their clients - who under MiFID will become competitors. Consolidation across the exchanges will lead to a price war.
PM: A fundamental shift in the way trading is carried out. Machines not humans will be trading liquid instruments and the reaction time needed to hit the best price in the market will be milliseconds, not the seconds that humans need. The market will get faster and will need faster chips to drive it.
PJ: 2007 will be the most memorable of our professional lives. It’s not often that all the participant roles are up for grabs: market data vendors, exchanges, regulators and the investment firms. The costs have got to come down but getting them down is a huge challenge. Ultimately the customer is going to win. There will be some big new pressures from customers and this will change the game.
Filed under: Issue 1 - Spring 07 | No Comments »
Loading ...