Introduction

Following the discussion of FAST, this article continues on the important subject of data. Accurate, low latency and absolutely reliable data is a necessity for automated trading. Unfortunately, it is usually one of the most difficult and expensive things to get right. A modest set-up with one or two direct feeds and associated broad band leased line connections can easily result in a annual bill in excess of six figures. Costs of this magnitude are prohibitive for most self directed proprietary traders and smaller hedge funds. This article looks at some of the options and technology that is available to the small buy-side trader.

Hosted solutions

For true black-box traders the problems can be neatly avoided by using hosted solutions. The trader simply rents space  in a server room with all the appropriate feeds and market access already wired in. The costs of connectivity are covered by the host but these costs can be split amongst many traders using the facility. Hosted trading has been available for many years through brokers like Genesis Securities, however, it is becoming an increasingly common solution. It is now possible to get hosting facilities that are located in a country equidistant between two other countries in order to minimise latency for arbitrage operations.

The disadvantages to hosting are a lack of hands on control.
If something goes wrong and you need to get access to the equipment in an emergency then the fact it is not on site can be a problem. There are also many automated systems that require monitoring and must have the facilities to allow a trader to take control when the unexpected happens. As soon as a system starts to include a significant amount of GUI support and manual trading facilities a hosted solution becomes much less attractive.

On-site point-to-point leased lines

If the decision has been made to keep the equipment on-site then the issue of connectivity needs to be addressed. For most professional trading operations the traditional solution has been point-to-point leased lines or ethernet connectivity. The advantages are fairly compelling:

1) Lowest latency and minimal transmission jitter.
2) Dedicated bandwidth.
3) Managed end-end allowing stringent service level agreements (SLA).

There are companies that specialise in offering ultra-reliable point-to-point connectivity to the financial community such as BT Radianz. Connections with 99.9999% availability can be obtained for critical trading operations.

The disadvantages are cost and inflexibility. Point-point connections are usually charged on a per-distance basis. If the remote site is a significant distance from the trading office then costs can quickly become prohibitive. This is particularly true if the trading office is outside of a metro-area where telecom companies are subject to less competition. In addition to line rentals, there are also often significant installation costs and long contract periods both of which are a disadvantage if you wish to change service providers or move offices.

The Internet

The professional financial services industry has generally taken a fairly dismissive view to using the internet for critical trading data. This is partly due to the fact that the internet does not offer any guarantees of connectivity but also possibly due to ignorance about what can now be achieved with the appropriate technology. To date, almost all internet based trading services have been oriented towards the retail investor and practically all of these are not adequate for reliable automated trading. Despite current prevailing attitudes and bad use of technology, the internet deserves serious consideration as a means of delivering trading data to small hedge funds and proprietary traders.

Some of the negative attitude may stem from the early days of the internet. When many companies first started looking at the internet, transatlantic ping times of more than 500ms were common and the world-wide-wait was an every day reality. However, the last two years have seen dramatic improvements in the internet to the extent that the world-wide-wait has all but gone and transatlantic ping times of around 80ms are normal. Recent times have seen the introduction of very affordable IP-leased lines. These provide a dedicated connection to the internet back-bone. As a result the user gets a guaranteed bandwidth connection into a giga-bit network. The use of IP-leased lines offers a viable, flexible and cost saving alternative to the traditional use of point-point connections but still requires the appropriate technology to deliver data.

VPN (The wrong technology)

Virtual-Private-Network (VPN) has become very widely used as a method of connecting remote offices and mobile workers to a central office. It provides a way of connecting the networks of two remote sites whilst ensuring security as the traffic is passed across the internet. Security is achieved by encrypting the data and the this is usually done with a protocol called IPSec.

It has also started to find some use within the trading community and some companies now offer streaming data over the internet via VPN. However, the technology was never intended to support this type of application and there are some important disadvantages to the technology:

1) VPN is not resilient. It is set-up between two IP addresses and hence requires that the internet service providers (ISP) at both ends of the link must not fail. It effectively introduces a single point of failure.

2) It is difficult to manage. Each user requires an individual sub-net. If there are thousands of users then thousands of sub-nets are needed. If an error is made then there is the danger of joining customer networks together giving rise to potential security risks.

3) It adds unnecessary overhead. Exchange data is basically public domain data and so does not warrant heavy encryption. As a result the encryption used by VPN has no benefit but adds latency, processing and bandwidth overhead.

The issue of resiliency is the fundamental flaw with using VPN for automated trading.  A typical ATS will require availability of at least 99.999%. The best ISPs will only offer service level guarantees of 99.9% or at best 99.95%. However, this issue can be addressed with clever router technology. A company called Xrio offers a range of load-balancing routers that allows a VPN to be established using four or more  ISPs, two at either end. The data is load-balanced in a mesh between the ISPs at the IP level. This process is transparent to the TCP and IPSec protocols that are established on top of it .
The result looks like a single resilient connection such that any one ISP at either end can fail and data will continue without any interruption. The addition of ISP redundancy in this fashion allows an availability of 99.9999% to be achieved with inexpensive internet connections. The principle disadvantage with this system is that is requires expensive router boxes at both ends of the link.

Other technology

Aside from VPN most internet data is transmitted in a range of proprietary formats over a basic internet connection. Most of these are designed as a minimum cost solution for getting trading quotes to retail investors.
However the market data company DTN, have recently introduced a new service called NxCore which moves the technology in the right direction.

NxCore has been aimed at the professional trader and hedge fund industry. It allows large content sets to be delivered across the internet using a compression algorithm whilst removing the need for expensive dedicated leased-line connections. For small funds the potential cost savings are clearly attractive. The technology also includes fault-tolerance features. If it is interrupted it will automatically back-fill data when the connection is restored.
This is a useful feature but a technology that is designed to be uniterruptible is better than one that just fails gracefully. The service does not support watch lists and will only allow entire market datasets to be enabled or disabled. This makes it inefficient for traders that require a select set of instruments  across a range of markets. In particular, this makes it less attractive for futures traders who might require data for a small number of futures on a range of exchanges and will find their connections slowed down by lots of unwanted options data.

The trouble with TCP

Current internet based data services almost exclusively use a TCP/IP connection in one way or another. The internet protocol (IP) ensures data is correctly routed from source to destination. However, it does not ensure data packets are received in the correct order or are not lost or damaged.
The job of ensuring data is complete and received in the correct order is done by the transport control protocol (TCP).

TCP provides a general reliable connection for transferring data across the internet. However, it was not designed specifically for streaming data with low latency. To understand some of the problems it is necessary to understand the basics of how TCP works.

All packets transmitted over a TCP connection are appended with a sequence number. The receiving end uses the sequence numbers to ensure received data is correctly ordered. If a packet is received with a later sequence number then this is put in a buffer until the prior packet is received. For example if sequence numbers are received as 1,2,4,5,3, 6 then packet 4,5 would be stored until packet 3 arrived. The data can then be assembled and passed to the receiving application in the correct order. This is the first problem with TCP, namely that data tends to be held up by the slowest to be received packets. This feature has a direct impact on latency.

The second issue to understand is what happens if a packet is lost or damaged. Every correctly received packet is acknowledged by the receiver which sends back an ACK to the transmitter. If the ACK is not received then after a timeout the data is retransmitted. At a very minimum the timeout must equal to the round-trip time (RTT) to allow data to be received and an ACK to return. It will take a further half RTT to send and receive the replacement packet. Hence, a lost or damaged packet will result in a minimum tripling of the delay to receive that packet. In practice, the timeout period must be significantly longer than the link RTT to allow time for the ACK to be received and ensure spurious re-transmissions do not occur too often. Hence, link latency is sensitive to packet loss.

The right technology

So far we have discussed some of the disadvantages attendant to existing solutions. Current standard internet protocols do not meet the needs of the financial service industry in allowing data to be delivered across the internet with very high reliability and the minimum possible latency. In the next article, an architecture and protocol that achieves these aims will be introduced.

Conclusion

The article covers some of the connectivity options available to the trader. True black box traders will likely find a hosted solution is optimal whereas well capitalised institutions will be able to afford the costs involved in direct feeds and leased lines. However, for small buy-side traders there is currently little in the way of services that fall within budget. The internet offers a practical path to fill this gap. However, existing protocols are not optimum for high reliability, low latency data transmission.