IBM's mini computerslack thereof 834

Latency is a problem no matter what the physical layer on the interface. So SCSI has command tag queues and scatter-gather capability, to minimize turnaround on the data bus. Transitioning between states on SCSI is relatively expensive, but once the data transfer state begins it's fairly efficient at transferring lots of data quickly. You can see the influence on IDE SATA drives as they are now moving to the same command structure as SCSI: packets, command queues, and (an improvement in thruput over SCSI) one channel per drive.

To some extent propagation time is hidden by seeks. Since there are three or more orders of magnitude in delays between the two, the time for the command packet to get to the drive isn't that big a deal. In some cases the command is going to a cluster or raid controller anyway, followed by multiple command streams to several drives in parallel, so the propagation time may not represent much in the overall end time.

The original 5MB-sec SCSI narrow bus did have timing restrictions that reduced some devices to 2-3MB-sec if there was a lot of bus state activity, even though there were no cable length issues. Jack Peachicken