The Securities Technology Analysis Center (STAC) recently announced the creation of a new benchmark council that includes some of the leading securities firms such as JPMC, Citigroup and HSBC. The new council will establish benchmarks in three areas:

• Market data: benchmarks based on workloads such as direct exchange-feed integration, market data distribution, tick storage and retrieval, etc.
• Analysis: benchmarks based on workloads such as trading algorithms, price generation, risk calculation, etc.
• Execution: benchmarks based on workloads such as smart order routing, execution-related messaging, etc.

“STAC Benchmarks will measure the performance of software such as market data systems, messaging middleware, and complex event processing systems (CEP), as well as new underlying technologies, such as hardware-based feed and messaging solutions, hardware-based analytics accelerators, compute and data grid solutions, InfiniBand and 10-gigabit Ethernet networks, multicore processors, and the latest operating system and server technologies.”

Benchmarks in complex event processing, huh ? That will be interesting. Will it be based on specific sets of use cases ? Will it give us insight into the hype and myth of the various CEP vendors’ proclamations of processing gazillions messages / sec ? Will it tell us what happens when you try to scale these products ? I wonder what the various CEP vendors think about this …?

Some FAQs here.

I saw Guy Churchward, VP Engineering at BEA, give an interesting keynote speech (including a Shakespearean intro) on BEA’s new LiquidVM product at BEA World in San Francisco the other week. Guy runs BEA’s product strategy and development for technology in the virtualization/utility computing, event driven architectures (real-time and complex event processing), portals and JVM (JRockit) domains.

LiquidVM is yet to be released as a GA product — but it is essentially JRockit with various extensions that allows application to be run in a virtualized environment on the Hypervisor of the host machine (ie without the OS). The benefit is improved performance and utilization, better compression of workload (e.g. supporting massively more market data feeds in the data center), and cost reduction.

The keynote is now available here, and whitepaper here.

Lab49 has been working with BEA on their new Weblogic Event Server (WLEVS) product in the complex event processing space. As part of this effort, we have been building a demo of how one could automatically reprice a portfolio of fixed income instruments against streaming market data. In addition to WLEVS, we used Quantlib’s C++ libraries and Lab49’s own market data simulator, with the front end built in WPF.

A couple of us from Lab49 attended the BEA World Conference in San Francisco last week and I presented the demo as part of the “Introduction to Event Server” session.

Click on the image below to view a screencast of the demo .. (this is a .wmv version for the moment..)

We were able to run the demo pricing 400 bonds/sec, with 4.6ms latency, on an Intel Quad processor - the folks at Shuttle helped us with their latest snazzy new XPC box. We also developed a cool WPF visualization of the Event Processing Network where we can automatically generate XAML from the underlying EPN configuration which we then data-bind to the event server’s performance monitoring meta-data.

I will put the demo and more information on the Lab49 website, and hopefully we will get it onto BEA’s Dev2Dev portal soon…