SUMMARY: FWD max devices

From: Kurt Ruthmansdorfer <kurt_at_sdss.fnal.gov>
Date: Thu, 9 May 96 10:11:21 -0500

Thanks to all who reponed to my question about the max number
of devices on PCI FWD scsi buses under DX. All text follows with boiler
plate cut out.
My conclusions:

- The limit is 7 peripherals even under DX 4.0. This is very disappointing.
We should have put a requirement in our bid. I wish I had known
about this group when we did our bid.

- A suggestion to put the controller at id 8 allowing 8 peripherals (0-7)
is intriguing and I may try it. With 8 devices in a chasis it would
make cabling easier.

- Suggestions from Nick Hill about configuring for optimized performance
are very useful.

- A suggestion to use a raid controller is out of the question. Cost per
GB is a major concern.

- The man rz page says " four bits of the minor ... specify the target ID "
which leads me to believe the data structures and existing minor device
number support wide devices. Maybe there is hope.

- There is a warning about a lot of elite 9 failures. We have hundreds
of them and they are quite reiilable.


=========================
Responses

>From alan_at_nabeth.cxo.dec.comThu May 9 09:08:27 1996
Date: Wed, 08 May 96 09:42:07 -0600


        Digital UNIX currently only supports 8 targets per bus;
        typically 1 controller and 7 devices.

>From alan_at_nabeth.cxo.dec.comThu May 9 09:08:46 1996
Date: Wed, 08 May 96 10:35:19 -0600


        Consider this 3rd hand unreliable knowledge... I've
        heard that the original V4.0 plan had lots of stuff
        in it. One of things included was support for Wide
        SCSI *addressing* as well as wide transfers (*). But,
        in the modern tradition of any large software product
        the volume of stuff going in made the release too big.
        In order to get a reasonable schedule, stuff was slipped
        to later versions. As far as I know wide addressing was
        one of those things.

        I think some of the data structure changes were made to
        support more than 8 targets per bus, but the rest of it
        doesn't seem to have gotten in. When one considers that
        Digital UNIX and ULTRIX SCSI drivers have been assuming
        8 targets per bus for years, it is a considerable change.

        Just one example is the number naming convention. You
        find the device number from:

                (bus * 8) + target

        That doesn't work on 16 target buses...

        (*) We've had wide transfers since the KZTSA was supported
        in V2.0.

>From chu_at_musp0.Jpl.Nasa.GovThu May 9 09:09:18 1996
Date: Wed, 8 May 1996 10:25:17 -0700
From: Eugene Chu <chu_at_musp0.Jpl.Nasa.Gov>

DEC OSF currently does not have provisions to address SCSI IDs beyond 7,
even though Wide SCSI should allow address up to 15. One trick I've
seen, but have never done myself, is to set the controller address to
15, since it never sends a packet to itself. This leaves all 8 IDs on
the bus available to devices.

eyc

>From johns_at_totaltec.comThu May 9 09:10:03 1996
Date: Wed, 8 May 1996 15:31:07 -0400
From: John Seel <johns_at_totaltec.com>


At 11:17 AM 5/8/96 -0400, you wrote:
B

Kurt,

To the best of my knowledge, Digital Unix still only supports 8 scsi ids on
a scsi bus even on wide scsi. One being used for the controller (typically
id 7) leaving 0-6 for disks or tapes (or other controllers if dual -hosted).

Hope this helps.


/--------------------------\
| John Seel |
| Total Tec Systems, Inc. |
| (908) 906-6500 ext 185 |
| Fax (908) 906-9808 |
| "johns_at_totaltec.com" |
\--------------------------/

>From cherkus_at_UniMaster.COMThu May 9 09:10:15 1996
Date: Wed, 8 May 1996 16:40:29 -0600 (EDT)
From: Dave Cherkus <cherkus_at_UniMaster.COM>

Currently, DU only supports 7 devices per bus. It is being looked into,
but it is not fixed in v4.0.

-- 
Dave Cherkus ----- UniMaster, Inc. ----- Contract Software Development
Specialties: UNIX TCP/IP X OSF/1 AlphaAXP AIX RS/6000 Performance ISDN
Email: cherkus_at_UniMaster.COM  Tel: (603) 888-8308  Fax: (603) 888-4598
Live Free or Die - New Hampshire has 3 seasons: ice, mud and black fly 
>From cherkus_at_UniMaster.COMThu May  9 09:10:46 1996
Date: Wed, 8 May 1996 16:41:08 -0600 (EDT)
From: Dave Cherkus <cherkus_at_UniMaster.COM>
I forgot to add that many people would use an HSZ40 raid controller
to solve the problem.
-- 
Dave Cherkus ----- UniMaster, Inc. ----- Contract Software Development
Specialties: UNIX TCP/IP X OSF/1 AlphaAXP AIX RS/6000 Performance ISDN
Email: cherkus_at_UniMaster.COM  Tel: (603) 888-8308  Fax: (603) 888-4598
Live Free or Die - New Hampshire has 3 seasons: ice, mud and black fly 
>From chris_at_lagoon.meo.dec.comThu May  9 09:11:17 1996
Date: Thu, 9 May 1996 18:25:15 -1000
From: Chris Jankowski <chris_at_lagoon.meo.dec.com>
Kurt,
I think that DU V3.2C will only recognise 7 devices per bus.
I am not sure if DU V4.0 extends that to 15, but probably yes.
Regards,
Chris
 +-+-+-+-+-+-+-+ Chris Jankowski - Open Systems Cons.- chris_at_lagoon.meo.dec.com
 |d|i|g|i|t|a|l| Digital Equipment Corporation (Australia)   tel.+61 3 92753622
 +-+-+-+-+-+-+-+ 564 St. Kilda Rd, Melbourne 3004, AUSTRALIA fax +61 3 92753453
    "UNIX was not designed to stop you from doing stupid things, because
     that would also stop you from doing clever things." -- Doug Gwyn
>From NMH1_at_axprl1.rl.ac.ukThu May  9 09:11:46 1996
Date: Thu, 9 May 1996 9:54:32 +0100 (BST)
From: Nick Hill - RAL CISD Systems Group <NMH1_at_axprl1.rl.ac.uk>
Reply to: n.m.hill_at_rl.ac.uk
Kurt,
I have an 8400 with 6 PCI FWD scsi cards with a mixture of Seagate elite 9 and 
DEC rz29 disks on. A few things to note:
1) The PCI card can support 15 devices as per the FWD spec but DIGITAL UNIX 
still only supports 7 per scsi bus. This means you can only get 4*7*8.5
2) You will need to add an entry to the file /sys/data/cam_data.c for the 
disks and rebuild the kernel otherwise the performance to the disks will be 
poor. DU does not know about these disks and therefore acts conservatively by 
not useing the SCSI TAG queing to the disks. Turning this on makes a big 
difference. The enrty I use is:
{"SEAGATE ST410800W",17, DEV_RZxx, (ALL_DTYPE_DIRECT << DTYPE_SHFT) | SZ_HARD_DISK,
  (struct pt_info *)ccmn_rzxx_sizes, 512, DEC_MAX_REC, NO_DENS_TAB, NO_MODE_TAB,
  (SZ_BBR | SZ_REORDER),
  NO_OPT_CMDS, SZ_READY_DEF, 64, DD_REQSNS_VAL | DD_INQ_VAL,
  36, 64
}
but this may not be correct for your disks if they are fast wide differential 
directly onto the bus. You may need "SEAGATE ST410800WD",18  as the first two 
fields. I have the fast wide versions in storage works ba356 shelves hence no 
D on the end of the model number.
3) If you are using Advfs then setting the read and write block size (with 
chvol command) up to 256 from the default 128 also helps improve throughput
to the disks.
4) With some fairly basic testing I found that 3 disks can saturate the FWD 
SCSI bus. Doing large sequential reads from multiple disks in parrallel on 
one bus the max sustained bus transfer gets to about 17 to 18 Mb/s
Nick Hill
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
CISD, Rutherford Appleton Laboratory,                 Tel: +44 (0)1235-445598
Chilton, Didcot, Oxon, OX11 0QX, England.             Fax: +44 (0)1235-446626
N.M.Hill_at_rl.ac.uk            http://www.cis.rl.ac.uk/people/nmh1/contact.html
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Kurt Ruthmansdorfer WH6W 6.15  Fermi National Accelerator Lab; ms 234
kurt_at_fnal.fnal.gov             PO Box 500
708-840-8057  pgr 612-0866     Batavia, Il 60510-0500

Received on Thu May 09 1996 - 18:03:56 NZST