Previous Next Contents

3. Performance

Performance measurements are a bit complicated, because they depend on the VME device acessed. The measurements represented here are made with a VIP II VMEBus CPU as VME Device. The VIP II is a 10 year old 68010 CPU with up to 768K Dual-ported Ram that can be acessed in A24D16 mode, with no block transfers.

Reading this via mmap()-Access yields about 1.1 MByte per second, reading per read() and doing dma transfers yields 6 MByte per second, which is close to the maximum rate the VIP II can yield.

Using read()/write() access shows overhead due to the syscall involved, which also depends on wheter DMA mode is used or not. On the tested P120 board the overhead for both the lseek() and the read() together is about 15 microseconds for programmed IO and about 65 microseconds for DMA access. This gives (with the VIP II as VME Slave) a optimum DMA threshold of 34 bytes.

Other devices are not testet yet, since i have none to try with...


Previous Next Contents