With Red Hat, Novell - and now Intel, thanks to its $884m acquisition of Wind River - all crowding into the real-time Linux space, Concurrent has to keep on its toes and keep its RedHawk Linux, well, current.
With RedHawk Linux 5.4, announced Tuesday, Concurrent is slipping into Linux 2.6.31 and offering full compatibility with Red Hat Enterprise Linux 5 update 4. That's because RedHawk is a tweak on Red Hat, adding real-time extensions and other goodies cooked up by Concurrent to make it different from Red Hat's own Enterprise MRG real-time Linux.
With this update, RedHawk Linux is able to take advantage of all the power management and virtualization features of the latest enhancements to Intel's Xeon 3400, 3500, and 5500 processors, which sport the QuickPath Interconnect, as well as the most-current six-core Opteron 8400s and their HyperTransport interconnect. The real-time Linux from Concurrent is also updated for new chipsets from Intel and AMD, and provides support for PCI-Express 2.0 peripherals.
Concurrent says that it has also done tweaks in the memory subsystems so that NUMA clustering of processors on a motherboard using QuickPath and HyperTransport can be used to huddle real-time processors running on a particular CPU with the memory pages it needs to run those processes most efficiently. NUMA architectures can leave processors in one socket and memory in another, which slows down performance considerably. This may be just an annoyance with a generic operating system, but it can be deadly in a real-time OS.
RedHawk Linux knows how to replicate software libraries and modules around a group of processors and their memory, to allow a kind of parallel processing to boost performance of a specific code set if needed. RedHawk Linux 5.4 can span up to 48 x64 cores in single system image (that's eight six-core Opterons) - the prior release topped out at 32 cores.
This latest real-time Linux release from Concurrent also sports the latest video drivers from Nvidia and embeds the CUDA parallel programming toolkit as well, which means numerical calculations can be dispatched to Nvidia graphics cards within a system or to Tesla co-processors (basically outboard video cards with supercomputer ambitions) attached to the system through the PCI-Express bus. ®