Author Profile - Adam Woolhouse as head of sales of TS-Associates, developer of the TipOff low latency monitoring tool. Prior to joining to TS-A, Adam had held senior technical sales positions, most recently with Endace, provider of network packet capture probes.
TS-Associates, Trading Systems Associates plc, succeeds by partnering with leading financial markets participants to overcome the complex technology integration issues that relate to the deployment and management of middleware based financial markets applications. As a result of its focused technology consulting and through its close partnership with customers, TS-Associates is rapidly establishing itself as the leading developer of appliance based, hardware accelerated, real time middleware monitoring, trouble shooting and analysis products. TipOff, TS-Associates’ first product using proprietary hardware acceleration technology, has been purchased by leading US and European investment banks.
"Companies don't buy tools because they're short of tools, they buy to meet specific business needs, especially in hard times."
-- Adam Woolhouse, Head of Sales, TS-Associates
Monitoring any environment brings it challenges and real time financial market data is no exception. Global market data in aggregate comprises vast amounts of information – several million financial instruments some of which update tens or even hundreds of times per second. Delivery of this data requires multiple Gigabits per second of bandwidth. The data rates peak immediately following market open, significant news events and generally in volatile markets.
For automated and algorithmic trading platforms to make the right decisions and get access to the available liquidity, market data must be received with the least possible latency. It must then be distributed around the organisation in real time, and be managed reliably so it is not subject to re-transmissions; one of the main causes of latency. Many market data protocols use multicast UDP transport, which presents yet more monitoring challenges.
In boom times, with profits being made, budget is plentiful and upgrades can be made across the infrastructure. In times of economic pressure, more needs to be done to sweat existing assets. With money tight, the areas for improvement need to be pinpointed to maximise dwindling budgets.
Because of the high bandwidth requirements, proprietary market data protocols (middleware messaging) and the requirement to measure latency, specialist hardware and software is required to analyse and monitor such environments. Let’s examine why.
Firstly, high speed data does some strange things to traditional network interface cards (NIC) and the processors that capture it. Because data captured on a traditional NIC sends interrupts to the processor as it is taken into memory, higher and higher bandwidth sends more and more interrupts to the processor.
Eventually the processor and NIC cannot cope and packets are lost. In normal server environments this doesn't matter as the reliability layers in protocols take care of the dropped information, but in environments where loss of data is in itself of interest, it is counterproductive if the analyser is raising false positives by loosing additional data. Every packet must be seen, and be available for analysis.
Secondly, it may be necessary to write this data to disk for post-capture forensics. Here too, data rates over 1Gbps present challenges due to bus rates and disk controller limitations.
Thirdly, to measure the time of a packet’s arrival, accurate timing must be employed. Leaving the time stamping to normal CPUs and operating systems and the timing reference to NTP is not accurate enough.
For all of the reasons above, it is necessary to use specialist, optimised hardware to capture, write to disk and decode market data for analysis. Trading Systems Associates plc (TS-A) builds a product call TipOff. Firstly TS-A uses FPGA accelerated multiple port hardware to capture and pre-process market data. This transfers the data directly into RAM without interrupting the processor.
Secondly, TS-A uses hardware compression in TipOff to reduce this data, so that less disk bandwidth is required than capture bandwidth. This enables all of the data to be written to disk without requiring exotic and costly disk sub-systems. Thirdly, the FPGA capture cards timestamp the packets on ingress (to 10ns resolution) using highly accurate timing hardware. This enables data flows to be monitored to very high timing accuracy as they cascade through a market data system.
Once captured, and in order to give insight into the data captured, it is necessary to decode the messages so that sense can be made of the environment monitored. TS-A works with messaging companies and in some cases reverse engineers these protocols to provide this insight. Latency needs to be monitored at the network level but also at the message level to be able to see the experience of the applications, feed handlers, market data distributors, complex event processing systems and smart order routers.
Using multi-port capture cards and placing these ports at multiple capture points using network taps, latency can be measured through a system passively. By subtracting the time between packets and reconstructed messages at different points in the system, latency can be calculated.
A typical schematic of a monitored market data system is shown below, with the red rings showing the points of capture, with various, typically encountered protocols and message systems.
If monitoring market data systems has its own challenges, it also has common themes with traditional network monitoring;
• see every packet
• decode to the higher protocols to give insight
• analyse these protocols for faults, sub-optimization and capacity planning purposes.
TipOff from TS-A allows an organisation to do this.
For more information on TipOff visit our website or email sales@ts-a.com. TipOff is deployed in worldwide banking organisations in locations including New York, London and Tokyo. Adam Woolhouse is Head of Sales for TS-A.
Recent Comments