• Related Articles

• Data Latency - Why Speed is of the Essence

• 

  Distance Latency: Overcoming the constraints of Geography in FX

• 

  Winning the war against latency

Gaining competitive advantage with ultra-low latency FX connectivity

By Nicholas Pratt

The need for speed has long been a mantra of the equities market as aggressive traders look to extract any possible time advantage in the battle for the absolute best of the best in execution. But is ultra-low latency going to prove to be equally sought after in the FX market where, typically, time is less of an issue for the bulk of participants? Or will the market remain apathetic around the whole latency debate?

"I don't think latency will be quite the same sought-after trait in FX as in equities where there is a lot of hype," says Bob McDowall, research director, Europe, for Towergroup. "In FX although there is a lot of trading, much of it has a medium-long term focus where traders are taking a 1 to 3 year view of the world and responding to events. There are some specialist funds out there that are trading on the volatility of the dollar on a daily basis - these funds do exist but they are in a minority."

There is also the issue of whether the majority of trading systems are properly equipped to cater for high speed, high volume, low latency trading. "When you have so much data going through systems at such high-speed, you have to wonder how suitable most systems are. Apart from those systems that are bang up-to-date, most of them were never designed to take data at this speed. I have heard form clients that their systems experience a chemical reaction where the whole thing goes out of kilter. Firms engaging in low latency trading have to test their systems to ensure that their trading process is equipped for ultra-low latency. Otherwise the whole thing can implode and it can take days to detect," says McDowall.

Confusion over latency

For Jon Vollemaere, managing partner Europe and co-founder of FXecosystems, any confusion in the FX market over latency is because the concept is still very new to most participants. "An economist I used to work with would say 'it's
not important, until it's important'. This was normally in reference to indicators like non-farm payrolls but here I'd say that it is because it is still a relatively new thing for FX markets. Many participants realise that latency is now an important factor and they have to do something about it fast but are not quite sure who to turn to for advice or where to start. But the market does understand that less is more when it comes to latency."

Bob McDowall

"Firms engaging in low latency trading have to test their systems to ensure that their trading process is equipped for ultra-low latency. Otherwise the whole thing can implode and it can take days to detect,"

For any FX trader looking to invest money in exploring the 'less is more' concept, Vollemaere recommends they first undergo a comprehensive audit of the current speed or latency they get from their hardware, software and telecom topologies. It is then a case of addressing the different forms of latency - network latency, application latency and, first and foremost, distance latency. "It is simply the easiest place to start in latency reduction. For example, if you're trading transatlantic at 120milliseconds or more at a busy time and you can cut that down to a constant 69 milliseconds every time, then that's the first place to start."

When it comes to application latency, much depends on whether the applications were built in-house or bought off-the-shelf. "If it's an off-the-shelf application, then most shops are constantly looking at ways to lower friction in the process. The original design is normally focused on making the correct trade rather than the speed of that process. This is just about looking at it with a fresh set of (low latency) eyes."

Then there is the issue of network latency in all of its various forms - from serialisation to propagation to switching delays and more besides. "All of these can affect the overall total latency (OTL) figure an that is why it is important to look at the type of network you're currently using," says Vollemaere. "We work with several carriers internationally so we can produce a best of breed combination for each client. For example one carrier can now offer an 18 microsecond hop through their network connection points, where others are closer to 180. For some many hop clients that can make a material difference. Serialisation and Propagation delay is directly related to line speed, it's the denominator in the equation. If you're subject to these delays in FX, as most are, then you're using the wrong technology if you want to trade with low latency."

Another thing to avoid if you are interested in pursuing a path of low latency is any kind of 'legacy' technology, says Vollemaere, because they would not have been designed with low latency in mind and therefore will never perform at high speed. Legacy also affects the telecom links between major destinations as can bandwidth bottlenecks and data infrastructure deficiencies. But how big an issue is this for FX traders and how far are some willing to go to solve it?

"It's true that certain parts of the world are 'hot spots' such as who has the best link between Chicago and New York," says Vollemaere. "But just as important for a global market like ours, there are 'slow spots' like long range trading New York to Sydney for example or places that have extra factors like China or Dubai. But it does still come down to which way around the world you send the data. Only one carrier has the fastest route for each city to city link, and that's the one to use if your strategy is based on arbitrage or multi venue liquidity. The fastest route will always come at a heavy premium however, and for some clients it is worth the extra money. For others there are normally other routes at slightly less speed but far lower cost."

Justifying investment

If there is large spending involved, most firms will want some evidence that the investment is justified, so what tools are available for measuring latency? Vollemaere mentions a number of vendors - Corvil, Correlix and TS Associates - that offer such tools, adding that this has been an important development in the market. "Before you could ping a venue and that's about it, and to be fair that's all you really needed. Now you can constantly compare latency to venues on a dashboard, and even correlate that to the rates you get for a speed versus spread measure - now that has the power to produce some very interesting results.

Jon Vollemaere

"If Colonel Custer had listened to the Gatling Gun salesman it might have turned out differently for him."

"There are also vendors such as Trading Cross Connects, Streambase, Aegisoft (now part of Thomson Reuters) and Rapid Addition that have some great new weapons for the arms race. Banks, brokers, buyside and even vendors all have large gains to be made by using low latency technologies whether that's risk or cost reduction or lowering slippage and mistakes. If Colonel Custer had listened to the Gatling Gun salesman it might have turned out differently for him."

And on the subject of salesmen, what exactly should FX traders be looking for when enquiring about low latency services? "Well, there is a lot of cheap sales talk and claims that don't stand up," says Vollemaere. "Luckily we haven't seen that in FX yet, but I'm sure it will come. I would suggest picking a firm that understands the FX market and the nuances of how it works, that's key. You want a vendor that can offer more than one solution to the problem and who is up with the latest developments rather than just flogging legacy. You want one that is focused on this space and not wandering over from another market. You also need a partner who has some credibility when they speak and who is leading the change in this area."

Migration from Equities

Although Vollemaere and FXecosystem is firmly routed in the FX market, the leading low latency providers are coming to the FX market from the equities world, such as UK-based managed services provider Fixnetix. "Traditionally we have been focused on the equity markets where the challenges around latency first arose and where latency is measured in microseconds" says Alasdair Moore, business development director and founding partner. "However, the evolution of systematic trading in the FX markets means we are signing more and more dedicated FX customers".

"Traders have the choice of either building their own infrastructure or outsourcing to specialists like us." Moore argues that the fast-moving nature of the low latency industry necessitates constant upgrading which brings with it considerable cost. "There is no point getting your latency down to a certain level and then sitting back because what is considered low latency now will not be considered low latency in the near future."

Alistair Moore

"There is no point getting your latency down to a certain level and then sitting back because what is considered low latency now will not be considered low latency in the near future."

Distance latency is a particularly acute issue for the FX market, says Moore, because it is one of the only true global marketplaces. "As a company dedicated to eradicating latency across the technology stack we optimise any environment to achieve the optimal solution that can be achieved. Whilst there are a number of vendors who talk of what is possible, our customers, across multiple asset classes, are only interested in performance that is demonstrable where not only can we articulate different fibre paths but also show them in real-time."

More effective measures can be applied to end-to-end network latency says Moore who defines four specific areas - serialisation, propagation, network CPE and service transmission. "Serialisation occurs when packets are transferred onto the wire. The higher the bandwidth of the service, the less latency is incurred hopping on and off the circuit therefore raising the bandwidth of the line end-to-end for the service and decreasing the latency accordingly."

"Propagation is the result of the physical length the data packets have to travel down the Service line. Assuming most WAN services these days physically use fibre for its layer one services we can ascertain the additional delay. Light travels at 0.00548ms per kilometre over Fibre. Therefore if the service is located 100km away from the source this will add 0.548ms onto the end to end delay."

"Network CPE delay is the additional delay incurred by Ethernet Switches and Routers that translate and route the packets from end to end. Each in line device will have a latency cost and this can have a significant impact depending on the amount of devices you have in line. Service Transmission delay is the additional delay incurred by the underlying transmission networks that support the Network," he says.

End to end design

Direct DWDM CPE systems have the lowest latency and can incur sub micro second latencies on some of the newer devices hitting the market. Any service which translates signals from electrical to light and back again will also incur most latency as the transformation from light to electrical and back again has a latency penalty.

Moore comments that, "Whilst all of the above are key aspects to any low latency strategy Fixnetix focus not just on the network where Fixnetix contract with over 18 different fibre providers but also all software and hardware components. This enables us to deliver a complete end-to-end design of services using best of breed new technology and tune performance from network interface card (NIC) to NIC."

"To measure these latencies over and above the traditional IP Networking Management systems, specialised equipment and software is required, which we build into our service offering and enable us to wrap a guaranteed performance SLA around delivery. These systems enable us to measure micro second performance and are provided by our partners such as Corvill and Correlix."

When it comes to issues such as latency 'hotspots' in different global locations, Moore believes this is entirely a telecoms issue. "Invariably there are certain environments where there is no choice of providers so whatever problems reside these cannot be resolved without significant investment. In most cases hotspots or pinch points are where the telco has overloaded or oversold its own network without investing in greater bandwidth. These are relatively easy problems to solve for. When we provide customer connectivity it is dedicated bandwidth to that customer and we never share or sell the bandwidth from under the client. If you want 10Mb-1Gb you get 10Mb-1Gb and it is never contended."

BT Radianz is a network provider and has been addressing low latency for a number of years in the equities market but, says Jerry Brunton, director for marketing and communications at BT Radianz, it is only in the last 18 months that there have been any enquiries from the FX industry about reducing their latency. "This has mostly been from the specialist hedge funds that make their money by trading extremely quickly and they are now moving into the FX market. They've been through the whole latency issue in their equities trading and now they want to know how to apply it to FX."

Differences between Equity and FX markets

There are some differences in how latency should be addressed in the FX market in comparison to the equities market. There may still be a lot of one-to-one trading in FX but it is not the same as the single central exchange that we see in equities. There is also a difference in the complexity of the instruments and there is a much lower concentration of derivatives in the FX market. Most importantly, however, there are fewer high frequency traders in FX therefore there is less demand for low latency services and there is much slower progress in finding FX-specific solutions.

"One of the first things that the industry did in the equities market to address latency was to break down the trading process into different components - from the trading engine to the marketplace, to receiving a price, to executing - and to see what level of latency was evident at each step," says Brunton. "But the FX industry is yet to take this step, mainly because the majority of FX firms are not driven by high frequency trading. Instead it is the algorithmic traders and hedge funds that are leading this."

The first step that any low latency vendor has to take when engaging with a client is to ask them exactly what their idea of low latency is, says Brunton. "If it is two seconds then they will generally be fine with whatever system they already have. But if they are looking for sub millisecond levels of latency, then they will have to look for specialist products. The main objective is to reduce the virtual distance between the trader and the execution venue or counterparty. This can be achieved by a combination of co-location and faster components so that there is a communication path that is as straight and direct as possible, with low latency switchers and routers that eliminate as many hops as possible."

Geographical element

Is there a geographical element to latency? Are some less developed markets more prone to latency because of the basic infrastructure in place? Or do the major financial centres experience bottlenecks due to the sheer volume of trading? "The majority of trading is done in the major financial centres and that's where there is more effort to reduce latency. Having said that, it was only in the last 12 months there has been a fibre connection between New York and Chicago."

This delay was down to the legacy of the US telecom industry where AT&T had enjoyed a monopoly for decades before deregulation was introduced in the 1980s bringing different operators for each region. And it is similar situation in Europe where the financial centres of London, Paris and Frankfurt are as similarly spaced apart as New York, Boston and Chicago and suffer a similar lack of direct connectivity. "All of these financial centres want to be competitive and the key to that is connectivity," says Brunton.

For any firm that is seriously focused on low latency trading, they have to be looking at co-location. "It is the only way to mitigate the physical constraints of distance and the lack of international connectivity," says Brunton. However, he concedes, "if you are dealing with multiple venues, as in the case of FX, you have to consider how important it is to be co-located at each venue and how the expense weighs up against the potential benefits. In recognition of this dilemma, BT Radianz offers a proximity solution to clients. We have a number of hosting centres at various international locations and we can co-locate our customers using these centres rather than leaving them to co-locate at each one separately."

Alan Berryman

"If you select a provider that delivers the shortest fibre path and remove layers of legacy technology, you can lower latency fairly easily"

Technology partners

So what does Brunton believe FX firms should be looking for in an ultra-low latency technology partner? "At its most basic level, addressing latency is about finding the cleanest, fastest and straightest line between point A and point B but there's a lot to be said for knowing the market and knowing the business," says Brunton. "We connect to over 50 portals and we have people that have been in the FX trading business for a number of years. Whatever the industry, it is important for any vendor to recognise that it is about their customers' business more than it is about their own product. For example, we schedule our maintenance upgrades around the fact that the FX market is a 24-hour business across timezones. Maybe not every vendor in the FX market does that."

AboveNet is a leading provider of high bandwidth fibre optic networks in Europe and the United States. According to Alan Berryman, UK Sales and Marketing Director: "ultra low latency is becoming an important factor in a wider range of asset classes including FX, as the market finds equities trading increasingly fragmented and more difficult to secure consistently strong profits.

"Traditionally, it has been acceptable for trade execution in the FX market to be relatively slower. Participants are now demanding the same ultra low latency performance that they have come to expect when trading equities. And because their portfolios are increasingly cross asset and cross-border, traders are looking for high performance connectivity in order to support their need to be able to trade in multiple assets across multiple venues with transactions that can involve multiple currencies."

As a result, trading firms are increasingly looking to extend their FX requirement to include hedging and arbitrage strategies in tandem with equity trading. "Addressing low latency in the FX market is, however,
not simply a case of repeating what works in the equities market," says Berryman. "An FX trader will need to consider the fact that they will require multiple local as well as global connections, so it's important that their connectivity strategy includes a provider who can offer a range of routes with an abundance of direct fibre capacity. They should also look for evidence of on-net connections to all their key local facilities, banks and trading platforms ensuring that the highest levels of security and reliability are maintained with the appropriate Service Level Agreements in place."

The area of latency that is easiest to address and improve is the connectivity portion, says Berryman. "Latency is directly affected by the route distance and technology used. Simply put, if you select a provider, such as AboveNet, that delivers the shortest fibre path and remove layers of legacy technology, you can lower latency fairly easily."

Underlying architectures

Another provider of low latency network products is the UK-based Exponential-e. Head of financial markets Mark Cooper points out that although time is merely the distance between two events, not all point to point connections are the same. And he suggests that traders take the time to look at the underlying architecture of their trading connections. "For example, a complex routed, or layer three, infrastructure will introduce latency not there in what is called a single-hop or switched layer two architecture. Layer three users need to configure GRE Tunnels to create tunnels across the core network. These increase the complexity and therefore the overheads and latency.

"Single-hop layer two architectures in contrast allow messages to be distributed to all required servers as fast and efficiently as possible over the WAN. Single-hop layer two networks are essentially a simultaneous broadcast or multicast feature which can add significant value in a high-frequency automated trading environment. Within a financial trading environment the distribution of data, simultaneously between your servers connected over a Wide Area Network is critical when trading across venues in different geographical areas. Look out for network providers who can support multicast natively within the core, allowing you to control messaging groups without any further configuration overheads and ensuring your data is simultaneously delivered to all sites, allowing you to run many models concurrently."

Mark Cooper

"Look out for network providers who can support multicast natively within the core, allowing you to control messaging groups without any further configuration overheads"

Cooper suggests that firms also keep a close eye out for network bottlenecks and legacy infrastructure. "They are not always easy to uncover at the point of purchase. This is most likely to occur at the metropolitan area network level where backhaul network technologies being used aren't fit for purpose for use in such an unforgiving sector as financial trading. What's more many large network providers are still trying to sweat their installed asset so they sell IP VPNs. The trouble with layer three VPNs is that they introduce latency as part of the highly complex way they are engineered."

Conclusion

Ultimately traders must make up their own minds when looking at low latency network providers, however there are some general guidelines which can be followed, says Cooper. "Those network providers who can deliver an end-to- end Ethernet solution with an in-built multicast capability are best placed to address the latency issues associated with algorithmic trading.

"Latency figures often represent the round trip of a trading route. Some providers show figures which are so low that they seem too good to be true, and they usually are. This is because in many cases the figure quoted only represents a signal travelling one way, which of course is of little use. Also only those providers with bags of bandwidth should only be considered. Then you can have Gig-E hand offs as standard to minimise the delay caused by any serialisation on the network interfaces. Finally end-to-end SLAs and latency management tools should be offered to deliver guaranteed uptime and the lowest levels of jitter and packet loss."