I seriously doubt that the ZERO is a valid result though, because if that was the case, if it was truly ZERO, then the whole system would freeze whilst it was performing the test. ie mouse pointer would freeze, Workbench would freeze...I wouldn't be able to open windows or do anything whilst running the test. If it was ZERO then even navigating my drawers on Workbench whilst playing an aiff song would show a slow down in Workbench everytime it had to buffer....and I've never noticed any such slowdown.
It actually is close enough to zero. Another test, watch the Executive meter while copying a large file between partitions. On native A4K IDE, it eats nearly 99% CPU during the transfer. FastATA is no different, there's no controller chip with DMA access to accelerator fastram.
The system doesn't slow to a crawl because the OS can multitask. (This is why the RSCP benchmark mentions mouse movement - it will steal CPU time and might alter the result.) But to transfer continuously at it's maximum rate, it will utilize almost all CPU time. It has to, because the CPU is doing all the work of transferring! A good DMA SCSI controller will do the same while utilizing much less CPU, because the data transfer is handled by a dedicated processor which has direct access to fastram. (The CPU sets the parameters of the transfer, but doesn't do the work.)
BTW, I used to copy scsi hard drives one to the other using squirrelscsi and it wasn't all that bad....it was quite good! I do like SCSI! :laughing:
The Surf Squirrel doesn't count, AFAIK there is no DMA path from the PCMCIA port to accelerator fastram. So it's PIO only and equally CPU-intensive.
