Introducing guest poster Bill McLane, Sr. Product Architect for TIBCO Messaging, who is covering a new middleware announcement today. Messaging is of course *the* main transport for distributing events in event processing systems, and comes in various “shapes and sizes” for various different needs. The “need” covered by Bill’s post below is the area of extreme-low-latency (nanoseconds) work. And having seen some of the numbers, I have to say it looks impressive…
Today’s a pretty significant day for us here at TIBCO – one that has been several years in the making. We are announcing the launch of our new extreme-low-latency messaging solution – TIBCO FTLTM. In today’s electronic trading markets where even a microsecond of latency can be cost millions of dollars to firms, the latency arms race is on. TIBCO FTL is advancing the performance envelope to where no middleware has gone before. In internal performance benchmarks, we have achieved intra-host and inter-host latency that’s more than 40% lower than the lowest numbers reported so far. As appealing as these new world records are, breakthrough performance is just one side of the FTL story. What we are announcing today is a lot more than new records in the latency arms race – it is a whole new chapter in low latency messaging middleware design, where you don’t have to compromise on features to attain sub-microsecond performance. Let me explain what I mean by that.
The ultra-minimalist design approach to low latency
As customers, especially in Capital Markets, demand faster data distribution, the initial response by many messaging vendors has been to strip the messaging layer down and make it lean and mean. This approach delivered substantial improvements to latency for the early adopters that were looking to eek every last microsecond out of their application to application latency. However, the solutions started to look less like middleware and more like thin wrappers on top of the physical transports. Application developers were forced to build more functionality into their application layer and manage more of the data distribution layer directly in their application. This meant that the performance of the raw middleware layer looked good when tailored to specific benchmarks – but in real world applications, the developer had to take on the responsibility of building more features and functions into the application and managing these components outside of the middleware layer, causing the overall performance to suffer. This approach was all about sacrifice – if you wanted extreme low latency, the only way to get it was to sacrifice the features and functions that have become common place in typical middleware environments.
The TIBCO approach
When we looked at the emergence of extreme-low-latency era in messaging – where application latency is counted in nanoseconds and throughput is counted in millions of messages per second – we decided to take a completely different approach. Our priority with TIBCO FTL is to provide the highest performing messaging solution without making the sacrifices that other low latency middleware solutions require customers to make. Let me outline one of the several features FTL offers – centralized metadata management.
One of the ways FTL provides speed with flexibility is to decouple the metadata from the actual message payload and allow complete administrative control of this metadata without impacting the high performance message stream. This means that while other middleware solutions require application developers to build complex management into their application or infrastructure, FTL provides this natively out of the box with little impact to performance. While many applications can get by with sending only binary data, which FTL supports, it becomes necessary at some point to provide some context around the data so a structure or connotative format to a message begins to emerge. For many of the low latency solutions, this format can be defined but is not centrally managed which requires the application developer to build this understanding and parsing into their application layer. TIBCO sees this as a required component of the messaging layer and because of that it has a built in a centralized metadata management system that allows for applications to register interest in data models specific to a given applications needs.
This is just one of the several innovations we are announcing in FTL. One of the best ways to learn more, is to register for one of the High Performance Messaging Forums that TIBCO is organizing next month in London (April 19) and New York (April 28). These free events offer a unique opportunity to have in-depth conversations on High Performance Messaging – we’ll have technology deep-dive sessions as well as expert panelists discussing extreme-low latency messaging best practices and techniques. Hope to see many of you there, please come by and say hello!
