zhahai@gaia.UUCP (Zhahai Stewart) (11/12/86)
In article <1179@tekigm2.UUCP>, timothym@tekigm2.UUCP (Timothy D Margeson) writes: > A few more notes, ALL ST506 or ST412 disk drives use an 8 bit data path, an > most typically spin at about 3600 RPM. Data goes on and comes off the disk > at about 5 megabytes per second, WITH A GOOD CONTROLLER AND FAST MEMORY BUS! Actually, these drives and almost all 3.5" and 5.25" Winchester drives [EXCEPT THOSE USING RLL ENCODING LIKE THE SHUGART ST-238 WITH APPROPRIATE CONTROLLER] use an instantaneous data rate of 5 Mbits/sec (625 KBytes/sec). Counting the address and data marks, error correcting codes, CRC, and inter sector gaps, the sustained data rate cannot be much better than 1/2 MByte/sec within one cylinder. Most pc controllers cannot handle even that speed and skip one or more sectors between each read/write - this is called interleaving. Tim's point that a given hardware/software combination's best performance in terms of interleave factor varies from other combinations, and is best not assumed but tested. His system worked best with a 6:1 interleave (skip 5 sectors between accesses), which may be very dependent also on his use of VERIFY ON. (Probably not tho - since his reccomended test used disk reads only, not disk writes; be sure your test includes both if that write performance also matters to you). Also, all of these disks use a bit serial (not 8 bit) data path. Some controllers for the PC claim to be able to handle 1:1 interleaving, ie: no sectors skipped; the pc's DMA should be able to handle that; I do not know if a modification to DOS's disk drivers is needed. If there is some processing time needed for processing between sectors, this may not operate some applications as fast as a slower interleave... zhahai/HiSystems
timothym@tekigm2.UUCP (11/12/86)
In article <129@gaia.UUCP> zhahai@gaia.UUCP (Zhahai Stewart) writes: >> A few more notes, ALL ST506 or ST412 disk drives use an 8 bit data path, an >Actually, these drives and almost all 3.5" and 5.25" Winchester drives [EXCEPT >THOSE USING RLL ENCODING LIKE THE SHUGART ST-238 WITH APPROPRIATE CONTROLLER] ST506 and ST412 are current standards for hard disk interfacing, and were developed by Seagate Technologies, Inc. The ST-238 disk drive is a RLL mode drive manufactured and sold by Seagate Technologies as well, and uses the ST412 I/O convention. I was wrong on the 8bit data path, single bit data IS sent across the data cable, not the 8 bit path I described. SORRY for passing misinformation. >use an instantaneous data rate of 5 Mbits/sec (625 KBytes/sec). Counting the >address and data marks, error correcting codes, CRC, and inter sector gaps, >the sustained data rate cannot be much better than 1/2 MByte/sec within one Probably a better estimate of data rate than mine. >(Probably not tho - since his reccomended test used disk reads only, not >disk writes; be sure your test includes both if that write performance also My method DID test read AND write performance (COPY both reads, then writes to the hard disk, and does an additional read if VERIFY is on). Only the actual running of programs performs only reads, which I suggested as a secondary test. >Also, all of these disks use a bit serial (not 8 bit) data path. >Some controllers for the PC claim to be able to handle 1:1 interleaving, >ie: no sectors skipped; the pc's DMA should be able to handle that; I do >not know if a modification to DOS's disk drivers is needed. If there is >some processing time needed for processing between sectors, this may not >operate some applications as fast as a slower interleave... > zhahai/HiSystems I welcome the correction, but maintain that anyone putting a harddisk into their system needs to be aware of the implications of interleave - beyond the recommendations of the manufacturer. Thanks again to zhahai for setting me straight, and keeping me honest! -- Tim Margeson (206)253-5240 PO Box 3500 d/s C1-937 @@ 'Who said that?' Vancouver, WA. 98668 {allegra..inhp4..decvax..ucbvax}!tektronix!tekigm2!timothym