monia at am.shrmsg.shr.MTS.dec.com
Mon May 29 04:06:09 PDT 1995
In my opinion, a major throughput issue, totally unrelated to the SIP/SPI
standards, is bus dead time. ie. The time that a device owns the bus without
generating bus activity. An example (hypothetical) of dead time is when
firmware interacting with the SCSI bus controller gets diverted to perform some
unrelated task, such as interrupt service.
Tests by Digital some time ago indicated that although protocol overheads may
cause a significant loss in bus carrying capacity (in requests per second),
this loss can be minor compared to the dead time loss. Our measurements back
then indicated that such losses often exceeded protocol overheads by more than
a factor of ten.
The points here are twofold:
1. Dead time losses usually don't show up in the spec sheet and can't be dealt
with by the standards.
2. Implementors and chip designers need to understand the system and controller
chip overheads. There's no point in increasing the bus speed or shaving a few
microseconds off the protocol if bus throughput is dominated by other factors.
More information about the T10