On Jun  4,  2:17pm, Toru Nishimura wrote:
> Things are getting clearer to me.  Does your NEWASC kernel have any
> SCSI devices on PMAZ-A option card?  (People, we are discussing NEWASC
> driver, not NetBSD/pmax old ASC driver)

  I don't have any devices on the PMAZ-A card normally - it's too slow.
(Direct DMA access to memory is a big PLUS!).

> o With "12.5MHz" PMAZ-A;
> 
> asc1 at tc0 slot 0 offset 0x0: NCR53C94, 12MHz, SCSI ID 7
> scsibus1 at asc1: 8 targets
> probe(asc1:3:0): max sync rate 3.01Mb/s
> sd2 at scsibus1 targ 3 lun 0: <QUANTUM, EMPIRE_1400S, 1022> SCSI2 0/direct fixed
> sd2: 1335MB, 3053 cyl, 8 head, 111 sec, 512 bytes/sect x 2734996 sectors
> 
> IOZONE reports;
>         1233618 bytes/second for writing the file
>         1086044 bytes/second for reading the file
> 
> o And with "25MHz" PMAZ-A;
> 
> asc1 at tc0 slot 0 offset 0x0: NCR53C94, 25MHz, SCSI ID 7
> scsibus1 at asc1: 8 targets
> probe(asc1:3:0): max sync rate 6.25Mb/s
> sd2 at scsibus1 targ 3 lun 0: <QUANTUM, EMPIRE_1400S, 1022> SCSI2 0/direct fixed
> sd2: 1335MB, 3053 cyl, 8 head, 111 sec, 512 bytes/sect x 2734996 sectors
> 
> In this case IOZONE reports;
>         1245339 bytes/second for writing the file
>         1081006 bytes/second for reading the file
> 
> There is no large difference between kernels.  Difference would be
> 'longer timeout', exclusively?

  With a 1.2Mb transfer rate, you won't see any difference on the CPU side
of the transfer.  To see any real difference, you would have to measure the
transfer rate on the SCSI bus.

  If the PMAZ-A is configured for a 25Mhz clock, it will try to set the
sync period to get a 5Mb transfer rate, but because the clock is really
12.5Mhz it will really only be transferring at 2.5Mb.  If the sustained
transfer rate of the disk is slower than 2.5Mb, this would have no affect
on the actual transfer rate.  If the disk can sustain a faster transfer
rate, you would be unlikely to see much difference:  the overhead of
copying to/from the PMAZ-A buffer and all the other I/O overhead in
doing the SCSI operation will mask any difference.

Michael

-- 
Michael L. Hitch			mhitch@montana.edu
Computer Consultant
Information Technology Center
Montana State University	Bozeman, MT	USA