SUMMARY: Common SCSI myths

From: Alberto Ferrari (FERRARIA@mz.astra.com.ar)
Date: Wed Apr 23 1997 - 17:05:59 CDT


Real good answers!!.
(1)"Nicholas R LeRoy" <nleroy@norland.com>
(2)Ric Anderson <ric@rtd.com>
(3)Audrey Taylor <Audrey_Taylor@compuserve.com>
(4)Ray Trzaska <rtrzaska@uk.mdis.com>
(5)Jim Harmon <jharmon@telecnnct.com>
(6)Michael Kohne <mhkohne@moberg.com>
(7)Mark Bergman <bergman@phri.nyu.edu>
(8)Michael Pearlman <canuck@caam.rice.edu>
(9)Sean Ward <seanw@amgen.com>
(10)Bismark Espinoza <bismark@alta.Jpl.Nasa.Gov>
(11)David L. Markowitz <david.markowitz@litronic.com>
(12)Rudy Yu <ryu@anthem.com>
(13)"Matthew Stier" <mstier@hotmail.com>
(14)Tom Mornini <tmornini@infomania.com>

Thanks a lot to all who replied.
Specially appreciated were answers from Ray (4) and Jim
(5)

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
> a) is there any advantage in using differential SCSI
> interfaces over single-ended ones - asuming your
controller matches?
> I guess you could reach farther with differential, but it's
only
> an assumption
(1) That is correct. I'm not sure about with Ultra SCSI, but
with SCSI-II,
bus length was limited to 6mt, and up to something like 25mt
with
differential. It's difficult to find devices that do differential,
in general, but you can buy differential -> non diff
converters.

(2) I've got some differntial runs of 50 feet. I *think* the limit
is 125 feet
(4) differential scsi gives longer distances ( 25 metres
instead of 5 metres of total cable
   run as a rough estimate ) no appreciable differnce in
speed unless you are
   being plagued by noise on the single ended type.

   Differential lines are in pairs, with one cable having a
positive signal,
   and the corresponding pair having a corresponding
negative signal - this
   means that beyond a very limited radius the signal
appears as zero. in single ended
   the signal cables are sepated by grouded wires. In any
cable the signal pulse
   gradually gets smeared out, and this smearing is less
than in the singlew ended case.

   Also external noise will shift the signal either positively or
negatively. this shift
   in a single ended cable can either create a mark that
wasn't origianlly there, or
   remove a signal that sould exist. In differential cables the
important thing is
   the difference between the negative and positive going
pulses - and since both will be
   shifted in the same direction by the same amount, the
external noise is practically
   eliminated.

(also from Ray)
SunSolve Document infodoc/14075
....

SCSI Primer

_______________________________________________________________________________

        (non-fast) fast fast/wide
        ================== =====================
=====================
        SCSI-1/SCSI-2 SCSI-2 SCSI-2
        <= 5MB/sec throughput <= 10MB/sec throughput <=
20 MB/sec

_______________________________________________________________________________

        narrow wide
        ================== =====================
        SCSI-1/2 SCSI-2 (?)
        8-bit 16-bit
        8 address lines 16 address lines
        up to 8 devices, up to 16 devices,
        id 0-6 targets, id 0-6 targets,
        id 7 initiator id 7 initiator,
                                        id 8-15 targets
        50 pin "A" cable 68 pin "P" cable

   In multi-initiator systems, the scsi id 6 is usually the
second initiator,
   as that has the second highest priority on both the narrow
and wide
   buses.
_______________________________________________________________________________

the extra pins give both extra addressing lines which are
just ignored by, because they
are not present in, narrow scsi devices/controllers

....
        single-ended differential
        ================== =====================
        SCSI-1/SCSI-2 SCSI-2
                                        superior signal quality
        <= 6m total cable length <= 25m total cable length
_______________________________________________________________________________

  Any combination of narrow/wide with
single-ended/differential
  with non-fast/fast is possible.

  Fast/wide combo yields up to 20MB/sec throughput.

(5)Differential SCSI is a big improvement over SE
(Single-Ended).
Yes, you can reach further, because the devices are better
able to
handle data traffic control, and has higher tolerance of data
integrity.

The length difference is approx 20 feet for SE moving up to
25 Meters in
Diff. That meant the ability to add full-size expansion
subsystems onto
a SCSI chain. With SE, the constraint meant you hac to
have all your
peripherals in the same box or cabinet as the SCSI
Controller. That is
highly inconvenient for a desktop user who needs more
than say--10GB of
storage.

Differential revolutionized the SCSI peripherals industry
nearly as much
as fast-wide. It actually created the possibilty of massive
RAID
architectures.

So, Diff gives you greater length, higher reliability, greater
capacity,
and --lest I forget to mention-- more intelligent SCSI
devices.

In fact, Differential SCSI devices could be considered to be
Sub-Master
Controllers instead of Slave controllers. Each device is
capable of
grabbing the bus without waiting for acknowledgment from
the host
controller, if the bus is not busy, and each device is
capable of
detecting a busy bus without having to wait for the host's
confirmation.

That was another contributing factor in the development of
Massive RAID,
and the qualifying factor in the extended bus length.

(6)Differential gives you better noise immunity, which
means less likelyhood
of problems with the bus. In very short runs, it doesn't matter,
but the
longer the cable, the more reliable this will make your
system.

(7)Longer cable lengths, less susceptibility to electrical
interference. Even with short cables, differential will be
better at
resisting interference. However, it is more expensive and
devices
are less common.

(8)Your assumption is correct, according to the SCSI
standard single ended
SCSI can go a MAX of 6 meters whereas differential has a
MAX of 25 meters.
Note:
1) these max numbers assume perfectly matched
components which does not occur in real life
2) FAST SCSI specifies a shorter MAX standard and
ULTRA even shorter)
                
(9)The only advantage of differential is longer cable length.

(10)Differential cabling reduces the amount of noise and
crosstalk that cables experience.

(11)Diff gives you longer buses with less chance of errors.

(12)(a) Yes, the major benefits of differential interface are :
        (1) Better SCSI signal over the SCSI bus (noise
immunity)
        (2) Longer cable

(13)Differential SCSI Interfaces can reach distances
'magnatudes' farther than
single-ended interfaces, and do it much more reliably.

(14)Correct. Length of cable is the only issue.
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
> b) Does putting a Narrow SCSI-II Fast device on a SCSI-II
> Fast-Wide bus (controlled by SCSI-II Fast-Wide card)
slow
> down overall bus performance?
(4) shouldn't slow anything down as long as the
terminations work - the last wide
   device sould self terminate the 'wide' part of the bus.
Some interference may
   still cause problems and in these cases a special cable
which terminaes the
   wide part may be obtained - or just get a separate
controller.

(5)Quoting from FWB's "Guide to Storage", by Norman
Fong Ch.3, pg 207:

        "Both Fast and Wide SCSI can exist on the same
bus. Combined,
         rates of 40 MB/s are possible. However,
implementing Fast
         and/or Wide SCSI capabilities may require silicon
changes in
         the SCSI Protocol chip. Because much
imformation will
         continue to be transferred in the eight-bit
asynchronous
         mode, fast and wide capabilities may not
significantly
         impact the I/O rates of your system unless you use
fast
         drives and CPU's."

It also suggested that if you are mixing Fast and Wide
devices, to put
all of one type on one side of the chain, and all the other on
the other
side of the chain using the controller in the middle.

Another point is that you may have to modify the connection
to install
Fast on a Wide bus by creating "partial terminators". Not
necessarily
simple, but the upside is that the author mentioned that
many wide
controllers will allow partial termination on the device.

I would ask you first, do the devices on the end contain
self-terminating circuitry? If they
do, you can terminate on the devices at the outside ends of
the chain. If they do not, you have
to terminate the ends of the chain itself.

(from subsequent message)
If the device is self-terminating, then the signal ends at that
device, for example: if you
have fast chain internally, with 4 drives, the one that is
farthest away from the controller on
the chain is the one to concider. If it has the ability to install
terminating resistors on the
device, and they are installed and enabled, then you don't
need another terminator on the end of
the cable itself, even if the end of the cable is a
pass-through connection to an external port.
If there is no way to terminate the last device in the chain
you MUST terminate the cable
itself. So, the end of the cable farthest from the controller
MUST have termination--as long as
there are any devices on the chain, either on the last device
itself, or the end of the cable
itself.

(7)I think (this may be another myth) that the overall bus
speed is
limited to the slowest device.

(8)Maybe, from a SCSI standard standpoint no but there are
two factors
*1) SCSI host adapters which have problems dealing with
the two different
device widths (this is rare today)
2) contention for the bus. The SCSI bus only allows two
           devices to utilize the bus at a time. Other devices
           wanting the bus must wait for the bus to be free.
Other
           things being equal a wide device will transfer the
data
           twice as fast as a narrow device lessening the
time
           the bus is occupied.

(9)No. Bus performance is the same for both narrow and
wide scsi.

(11)That depends on what you mean by "performance". It
will not cause
the wide devices to run in narrow mode. When the narrow
dev is idle,
the rest if the bus will behave as before.

(12)(b)(d)Yes/No. Make sure you do NOT mix
"Single-Ended" and
"Differential" devices. Technical speaking, your host
adapter
(20MB/s) can handle narrow device (10MB/s), but unless
you plug SCSI
analyzer on the SCSI bus you probably won't see big
difference.

 Basically, you can mix any "Wide" and "Narrow" devices
with "Wide"
and "Narrow" host adapter. ---- **** One thing you need to
be awared
of is "Terminator" ****.
If you were to run the narrow cable to the wide drive first,
and
then on to a narrow device, and terminate narrow device.
Problem!!
(because the narrow portition of the bus would be
terminated all the
way through, but the upper bus bits on the wide drive, altho
not
being used are unterminated)

(13)If the device is accessed often, and for any other FW
device farther down the
chain.

(14)As long as the termination of data line is handled
properly,
and the narrow devices are attached to the end of the cable
(farther out than the wide devices) the only degredation is
that the bus will be tied up during the slower narrow
transfers
for more time than it would be tied up during a same sized
transfer from a wide device.

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
> c) Is it wise to separate disks from other peripherals (like
> tape drives, CD-Rom drives and the like) in different SCSI
> buses? Why?
(4) I like to have the tapes on a separate controller, sun
recommends a maximum of
   two tapes on narrow or four on wide. I find that when using
tapes they do degrade
   scsi channel usage - so dont put heavily used disks on
the same controller as
   tapes.
(5)This would most depend on how you use the devices,
what they are, and
how many of them you have.

In a multi-channel RAID situation, you would best put an
equal number of
HD's on as many channels as you could, and hold a spare
channel for
other devices, like CD's, scanners, plotters, etc.

If you have a 2 channel system and you had 2 very large
HD's, one small
HD, and a CD, and you were going to use both big disks for
--say-- an
ORACLE Database, with the system disk being the smaller
drive, I'd put
one Big drive on each channel with the CD on one, the
smaller HD on the
other, so that the performance of the database was
marginally enhanced.

If you have 2 channels with a CD, Tape drive, and 2 HD's,
one a system
disk, the other your "applications" disk, I would put the two
HD's on
one channel and the CD and Tape on the other so that the
system Backups
would be better served. The SCSI channels (at least most
of the
multi-channel controllers I've dealt with) can buffer data
between reads
on one and writes on the other, giving better throughput on
writing to
the tape, and reading from the CD when installing new S/W.

The idea is to distribute the highest traffic data exchanges
between
channels.

(6)Well, a slow peripheral could soak up bus bandwidth
simply because it has a
limited transfer rate, and therefore bus transactions take
longer. It
really depends on the method of access, and whether the
system is using
disconnect/reconnect. If so, then it probably won't matter
much (system
starts transaction, disconnect, does another transaction with
another
device, and by the time that's done, the CD has finally
finished reading
the block off disk).

If I had multiple busses, I'd put the slow stuff all on one bus,
just to
limit it's chances of getting int he way fo the fast drives.

(7)See above. If I'm correct, than it won't matter as long as
the disks
are the slowest devices (unlikely). I doubt we'll see many
Ultra-SCSI cd-roms or scanners anytime soon.

(8) Load balancing and contention avoidance. Tape
drives and CD-roms
  tend to be slower than disks and thus a data transfer can
keep the bus
  busy longer. Consider the following scenario, backing up
a disk onto tape
  with a single bus, all traffic must go thru the cpu and thus
you are
  forcing sequential access. With multiple buses you can
have concurrent accesses.

(9) It depends. Disks with a lot of activity should probably
be on their own bus.
Otherwise, it doesn't matter very much.

(11)It is wise to balance your load as much as possible.
Dedicating buses
to one type of dev will only help if it is balancing the load,
ideally
by separating devices between which data is often
transferred, such as
a data disk and your backup device (tape drive). On the
other hand, if
you only backup at night when the system is mostly idle,
this may not
matter (unless you need high backup performance to finish
your backups
in the time allowed).

Similarly, try to spread mirrored disks and stripes across
multiple
controllers.

(12)Yes. SCSI protocol is a "buffered" protocol -- the SCSI
host
adapter in your system will negotiate the transfer speed
(not exceed
its max transfer speed) with each individual SCSI device on
the bus.
With more SCSI devices on the same bus, there is a
chance you are
going to saturate the SCSI bus. For best performance, try
"Wide" or
"Ultra Wide" interface for your overall performance
increased.

(13)The data rates for the two types of devices are
drastically different, and if
the slower devices are access a large amount of the time,
a significant amount
of bus time can be tied up talking to the slower devices.

(14)The more data paths to the computer the better. Slow
devices
tie up the bus longer than fast ones.

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
> d) May I put SCSI-II Fast/*Wide* devices on SCSI-II
> F/Narrow bus (controlled by a SCSI-II F/Narrow card)?
> I've mixed SCSI-II F/Narrow devices on Wide buses and
> they worked fine..
(4) yes, but you can only use narraw addresses on the
device, i.e. scsi id less than seven.
(5)See "b" above. Some FSt Wide devices will allow
partial termination on
the device. You'ld have to come up with the adapter for
50-pin to 68
pin tho.

        
(9)Yes. However you will not be able address the wide
devices above target 7.

(11)Yes, if properly designed, but they will only operate in
narrow mode
of course.

(13)Yes you may mix. However, it recommended that the
narrow devices always be
placed at the end of the chain. Once a chain goes narrow, it
stays narrow.

(14)Yes. The drives will negotiate down.

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
> e) where can I find more information (best if
> understandable) about these and other SCSI myths?
(5)A cool book I got in the "Macintosh Hardisk Toolkit Pro"
was the one
quoted above. ISBN 0-9651915-0-8, 29.95US. It's subtitled
"The
essential reference on computer storage technology". Of
course it
spends a good deal of time specifically extolling the virtues
of Mac's
but it has a comprehensive discueeion on SCSI and IDE,
including SE,
Diff, SCSI-2, and SCSI 3.

(12)http://www.cis.ohio-state.edu/hypertext/faq/usenet/scsi-faq

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>
> f) where can I find information about SCSI-III (SSA and the
> like)?

(5)See e.

Check out Adaptec's web site: http://www.adaptec.com/

I add: New Storage Interfaces - SysAdmin - Apr1997
>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>



This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:11:51 CDT