Making shared memory more efficient for multi-processor systems
Embedded computing firm Vmetro is proposing to simplify the development of multi-processor systems by changing the way data is transferred to the shared memory. The company’s FusionIPC processor-to-processor comms software handles both bulk data movement and message passing in a small footprint without requiring application involvement. “Similar to other existing approaches, FusionIPC is built upon a shared memory buffer model but differs in that it combines coordination of bulk data movement with messaging and signals without application involvement,” commented Mike Jadon, CTO of embedded systems at Vmetro.
“This reduces the complexity for application developers, but does not unnecessarily abstract them from the hardware and introduce additional overhead as would be the case with a conventional message passing based implementation,” added Jadon. The software supports a shared memory buffer (SMB) mechanism, “zero copy” bulk data movement within the system and message passing for status and control. FusionIPC is based on open standards with POSIX interfaces. Initially supported on the VPX MPE730 multiprocessor board, SBC731 single board computer, and HPE720 hybrid MPC8641D/FPGA board, FusionIPC utilises the Serial RapidIO (sRIO) fabric. It is also compatible with third party development environments including Freescale’s LTIB and Denx’s EDLK. Initially available on Linux 2.6, later distributions will also support VxWorks.