prindle@nadc.arpa (10/17/86)
The term interleave refers to the positioning of consecutively numbered sectors around the circumference of a disk cylinder. For example, a 2:1 interleave on a disk with 16 sectors per surface per cylinder would result in the following arrangement of sector numbers around that cylinder (n is starting sector #): |n |n+8 |n+1 |n+9 |n+2 |n+A |n+3 |n+B |n+4 |n+C |n+5 |n+D |n+6 |n+E |n+7 |n+F | Whereas a 1:1 interleave would be: |n |n+1 |n+2 |n+3 |n+4 |n+5 |n+6 |n+7 |n+8 |n+9 |n+A |n+B |n+C |n+D |n+E |n+F | As you can see, data transfer will be somewhat faster in the 1:1 interleave case since a block of 16 consecutive sectors can be read in one disk revolution instead of two. 2:1 interleave would be required in situations where the system timing could not guarantee that disk data could be DMAed into or out of memory fast enough to accomodate 1:1 interleave (if data transfer is slower than the rate at which data passes the head on the disk, the overall transfer rate would drop drastically, since a missed sector means having to wait a full disk revolution to re-read that data). A large enough buffer in the disk controller can compensate for transient drops in DMA transfer rate (such as when another DMA device is active simultaneously), but if the aggregate through-put (average) of the controller and processor DMA cannot handle 1:1 interleave, data transfer timing will be improved by going to 2:1. If the system can handle 1:1, then 1:1 will maximize throughput. An interesting side effect occurs when a pack is being formatted: when format- ting with 2:1 (or 3:1 or whatever) interleave, it is quite ordinary to format sectors in a simple consecutive order, since there is adequate time between sectors for the processor to issue the next format command to the controller. But in 1:1 interleave, if a formatting program tries to format consecutive sectors, formatting may take well into the next century; unless the processor is *very* fast, the disk will have moved too far from the end of the previous sector, and the controller will await another revolution to format the next sector! Thus formatting programs for 1:1 interleave may have to make two passes per cylinder per surface, once for the even numbered sectors, and once for the odd (or, if the processor is *really* slow, 4 passes). All of the above info is based on really BIG hard disks (250Mbyte and up) spinning at 3600 RPM. I don't know how fast these new mini drives spin, so some of the timing considerations above may not apply, but the general principle holds. Frank Prindle Prindle@NADC.arpa