IBM has released version 2.3 of its WebSphere MQ Low Latency Messaging offering, adding intelligent self management of messaging topics, assured delivery of messages, as well as a host of other functional and performance improvements, and previewing time synchronisation. The release follows news of the messaging product's adoption by Chi-X Global and Deutsche Boerse for their respective trading platforms.
Intelligent self-management within version 2.3 is achieved via a topic mapping service, which allows developers to create low-latency messaging topics without requiring them to understand low-level transport parameters. When applications use topics that are so created, they do not need to explicitly call low latency application programming interfaces, since the mapping service already understands to leverage them.
Topics are identifiers used within messages to uniquely identify types of data - for example, a topic might be defined by a sending application as "NYSE equities" allowing receiving applications to filter and process only those messages of interest. Assured delivery builds upon message store functionality introduced in version 2.2, allowing sending or receiving applications to confirm that messages have in fact been persisted to message store disk, to ensure that they are available for store and forward functionality.
The feature also offers the ability to set a maximum time limit after which messages are automatically written to disk. The combination of assured delivery along with the new positive acknowledgement feedback notification is essential to applications, such as order routing systems, which require both extremely fast execution and assurance that messages are not lost.
In addition to the above key functionality, version 2.3 also includes a number of enhancements at the transport layer to improve performance, add synchronous messaging, and provide better InfiniBand support.
Also new is latency monitoring at the reliable unicast level, providing statistics on internal latencies for unicast transmitters and receivers, and message queues, which can be used to balance resources consumed according to message rates. Intriguingly, the latest release also provides a "technology preview" of clock synchronisation functionality across multiple servers, providing a Coordinated Cluster Time (CCT) that can be used to accurately timestamp messages and so measure latency in a distributed environment.
The preview is supported only by LLM implementations running on Linux in an Intel x86 environment. Recent performance tests conducted by Chi-Tech, the technology services unit of Chi-X, demonstrated LLM's capability to underpin the technology components required by a liquidity venue.
In those tests, one million FIX order messages per second were processed with an average response time of 228 microseconds, in an environment that included IBM BladeCenter servers and InfiniBand networking.