Financial trading applications expose all the weakness of standard “servers”. The cumbersome network stacks that copy every byte of data many times and surround each byte of content with many bytes of overhead data. The rube-goldberg process schedulers. The memory management subsystems, designed on the basis of data from studies of applications used in university computer science departments in the 1970s. The file systems that have inherited block sizes from the PDP/11 era. The increasing speed of hardware, and the decreasing costs of huge memory and storage systems have compensated for flabby systems software for 20 years, but we are reaching a point where power consumption, cost, and exploding non-determinism cannot be cured by hardware alone. FSMLabs multiple kernel approach, which we used to great advantage in the embedded space for over a decade, provides a safe path towards radical improvements in system performance.
The basic idea of the multi-kernel approach is to use standard server system software as a utility that can run under control of a more advanced operating system. The technique we use on current technology hardware, which we sold to WindRiver for the embedded market and we continue to use in Enterprise, virtualizes a small part of the hardware so that the stanrdard OS can be preempted whenever timing critical applications need to run. FSMLabs Real-Time Management System (RTMS) essentially runs as a module on top of a server version of Linux allowing Linux applications and drivers to run unmodified, while preventing them from delaying timing critical applications. The next generation FSMLabs RTMS incorporates a new approach to multi-kernel computing and will be available in Q4 2007 and a following system will appear 2008. But the basic leverage of the multi-kernel approach is that a super high speed, low latency system environment is made available as a module so that investment in existing software is completely preserved even as order of magnitude performance improvements are made easily accessible. Four microsecond worst-case thread scheduling delay, 80 micro-second worst case round trip network packet routing, 1 microsecond worst case timing variation in cluster connected compute nodes - all of these are available now. Innovation has been made more difficult in systems software by success: nobody can afford to abandon hundreds of millions of lines of working software and known interfaces. The multi-kernel paradigm allows us to innovate ruthlessly in the new operating system level, while dragging along all of the standards based advantages of the older technology software.
.
Normal business rules apply
High speed/low latency business transactions require complete systems design approaches like that supported in FSMLabs RTMS.
FSMLabs launches its new web site as we exand our Enterprise Real-Time technology.
FSMLabs introduced light-weight virtualization more than a decade ago.
What is Real-Time?