RoB-in Buffer Manager -- Preliminary Timing Measurements

With code compiled at optimisation level 4 on a 33MHz i960RP I get the following crude timing measurements.
Time to index 10 single page events 69 microsecs
Time to index 20 single page events 139 microsecs

Not including time to receive message
Time to process decision record of 100 rejects 426 microsecs
Time to find data for an event AND set up DMA 15 microsecs

Including time to receive message
Time to process decision record of 100 rejects 558 microsecs
Time to find data for an event AND set up DMA 31 microsecs

The message passing overhead may be reduced but the other timings are more likely to get worse as more error checking and monitoring is introduced. These timings should in principle be improved by up to a factor of 2 by using the i960RD (a clock doubled version of the i960RP) but until we can benchmark it I wouldn't like to predict quite how the timings scale with clock speed.


GJC / gjc@hep.ucl.ac.uk