Suggestions for "the fastest" Q22/SCSI-interface?

[Tim Shoppa]

Dear list members,
I'd like your "educated opinions" upon which SCSI-adapter that would
be "the fastest" for the Q-22 bus... (if that can be achieved without
starting a "religious" fight...)
I've seen some indications that CMD's CQD 440 would be a candidate,
but on "bitsavers", I can only find a manual for CQD 420. What would
be the difference upon these two?
Reason for my questions: I've been given (privately) a Compaq HSZ-80
storage array, and it would be quite amusing to have a PDP (11/93)
entirely hooked up on this (besides an old alpha-station etc).
This storage is connecting using "Ultra SCSI-II" on
"Differential" (HVD) signalling, so diff (HVD) is sort of a
requirement of mine for the Q-bus interface I could spend some hobby
money upon getting this hardware, but as all hobby budgets, there is a
limit...
But, the "greedy" part of mine wants to find out which interface would
be "the fastest", just for the h... of it!  I'm primarilly running RSX
operating system, but others would be a fun alternative. With the
storage, I could have plenty of disc partitions for different OS:es!

I did (and published) some benchmarks of SCSI MSCP-emulating
controllers probably a decade ago. And indeed the CMD CQD 440 was the
winner. But they all beat the pants off a RQDX3!

Here are the peak data rates measured for read and write 64
blocks-at-a-time:


Read Write
---------- ----------
Andromeda SCDC 2.298 MB/s 1.131 MB/s
CMD CQD440 2.397 MB/s 1.525 MB/s
CMD CQD220 1.418 MB/s 0.882 MB/s
CMD CQD220A 2.088 MB/s 1.409 MB/s
DEC RQZX1 1.379 MB/s 1.097 MB/s
Viking QDT 0.846 MB/s 0.704 MB/s
DEC RQDX3 0.164 MB/s 0.161 MB/s

The benchmarks were done under RT11FB 5.7 doing 1, 2, 4, 8, 16,
32, and 64 block-at-a-time READW's and WRITW's to 16384-block
data files. A KDJ11B (PDP-11/73) CPU with 2 Megabytes of Clearpoint
non-PMI memory was used for the bencharmks. With the SCSI
controllers a Barracuda 7200 RPM ST15230N drive was used; with
the RQDX3 a RD52 drive was used.


Tim.

[Johnny Billquist]

Post by Tim Shoppa

Dear list members,
I'd like your "educated opinions" upon which SCSI-adapter that would
be "the fastest" for the Q-22 bus... (if that can be achieved without
starting a "religious" fight...)
I've seen some indications that CMD's CQD 440 would be a candidate,
but on "bitsavers", I can only find a manual for CQD 420. What would
be the difference upon these two?
Reason for my questions: I've been given (privately) a Compaq HSZ-80
storage array, and it would be quite amusing to have a PDP (11/93)
entirely hooked up on this (besides an old alpha-station etc).
This storage is connecting using "Ultra SCSI-II" on
"Differential" (HVD) signalling, so diff (HVD) is sort of a
requirement of mine for the Q-bus interface I could spend some hobby
money upon getting this hardware, but as all hobby budgets, there is a
limit...
But, the "greedy" part of mine wants to find out which interface would
be "the fastest", just for the h... of it! I'm primarilly running RSX
operating system, but others would be a fun alternative. With the
storage, I could have plenty of disc partitions for different OS:es!

I did (and published) some benchmarks of SCSI MSCP-emulating
controllers probably a decade ago. And indeed the CMD CQD 440 was the
winner. But they all beat the pants off a RQDX3!
Here are the peak data rates measured for read and write 64
Read Write
---------- ----------
Andromeda SCDC 2.298 MB/s 1.131 MB/s
CMD CQD440 2.397 MB/s 1.525 MB/s
CMD CQD220 1.418 MB/s 0.882 MB/s
CMD CQD220A 2.088 MB/s 1.409 MB/s
DEC RQZX1 1.379 MB/s 1.097 MB/s
Viking QDT 0.846 MB/s 0.704 MB/s
DEC RQDX3 0.164 MB/s 0.161 MB/s
The benchmarks were done under RT11FB 5.7 doing 1, 2, 4, 8, 16,
32, and 64 block-at-a-time READW's and WRITW's to 16384-block
data files. A KDJ11B (PDP-11/73) CPU with 2 Megabytes of Clearpoint
non-PMI memory was used for the bencharmks. With the SCSI
controllers a Barracuda 7200 RPM ST15230N drive was used; with
the RQDX3 a RD52 drive was used.

Speaking of which. Do anyone know what the difference between the
different CMD controllers are?

Johnny

[Tim Shoppa]

Post by Johnny Billquist

Post by Tim Shoppa

Dear list members,
I'd like your "educated opinions" upon which SCSI-adapter that would
be "the fastest" for the Q-22 bus... (if that can be achieved without
starting a "religious" fight...)
I've seen some indications that CMD's CQD 440 would be a candidate,
but on "bitsavers", I can only find a manual for CQD 420. What would
be the difference upon these two?
Reason for my questions: I've been given (privately) a Compaq HSZ-80
storage array, and it would be quite amusing to have a PDP (11/93)
entirely hooked up on this (besides an old alpha-station etc).
This storage is connecting using "Ultra SCSI-II" on
"Differential" (HVD) signalling, so diff (HVD) is sort of a
requirement of mine for the Q-bus interface I could spend some hobby
money upon getting this hardware, but as all hobby budgets, there is a
limit...
But, the "greedy" part of mine wants to find out which interface would
be "the fastest", just for the h... of it!  I'm primarilly running RSX
operating system, but others would be a fun alternative. With the
storage, I could have plenty of disc partitions for different OS:es!

I did (and published) some benchmarks of SCSI MSCP-emulating
controllers probably a decade ago. And indeed the CMD CQD 440 was the
winner. But they all beat the pants off a RQDX3!
Here are the peak data rates measured for read and write 64
                  Read           Write
                  ----------     ----------
Andromeda SCDC    2.298 MB/s     1.131 MB/s
CMD CQD440        2.397 MB/s     1.525 MB/s
CMD CQD220        1.418 MB/s     0.882 MB/s
CMD CQD220A       2.088 MB/s     1.409 MB/s
DEC RQZX1         1.379 MB/s     1.097 MB/s
Viking QDT        0.846 MB/s     0.704 MB/s
DEC RQDX3         0.164 MB/s     0.161 MB/s
The benchmarks were done under RT11FB 5.7 doing 1, 2, 4, 8, 16,
32, and 64 block-at-a-time READW's and WRITW's to 16384-block
data files.  A KDJ11B (PDP-11/73) CPU with 2 Megabytes of Clearpoint
non-PMI memory was used for the bencharmks.  With the SCSI
controllers a Barracuda 7200 RPM ST15230N drive was used; with
the RQDX3 a RD52 drive was used.

Speaking of which. Do anyone know what the difference between the
different CMD controllers are?

Speaking from deep memory:

the 2's are dual-height boards
the 4's are quad-height boards
If it ends with a 3, it's got the bulkheads for the Skunk-box (BA213)
style cabinet

I think there was a suffix to denote whether it spoke single-ended or
differential SCSI.

I think 4's had both single-ended and differential SCSI in some (all?)
cases... I think that's why they had two fifty-pin connectors, and
there were jumpers to select which connector was used. Geeze, it's
been too many years, like the Fuji Eagle jumper configs for sector
size that I used to have in my head too! Man, I could tear down five
broken Fuji Eagles and build back up 4 working Fuji Eagles in a
heartbeat at one time...

[Charles H Dickman]

Post by Tim Shoppa
I did (and published) some benchmarks of SCSI MSCP-emulating
controllers probably a decade ago. And indeed the CMD CQD 440 was the
winner. But they all beat the pants off a RQDX3!
Here are the peak data rates measured for read and write 64
Read Write
---------- ----------
Andromeda SCDC 2.298 MB/s 1.131 MB/s
CMD CQD440 2.397 MB/s 1.525 MB/s
CMD CQD220 1.418 MB/s 0.882 MB/s
CMD CQD220A 2.088 MB/s 1.409 MB/s
DEC RQZX1 1.379 MB/s 1.097 MB/s
Viking QDT 0.846 MB/s 0.704 MB/s
DEC RQDX3 0.164 MB/s 0.161 MB/s
The benchmarks were done under RT11FB 5.7 doing 1, 2, 4, 8, 16,
32, and 64 block-at-a-time READW's and WRITW's to 16384-block
data files. A KDJ11B (PDP-11/73) CPU with 2 Megabytes of Clearpoint
non-PMI memory was used for the bencharmks. With the SCSI
controllers a Barracuda 7200 RPM ST15230N drive was used; with
the RQDX3 a RD52 drive was used.
Tim.

Interesting. To add an additional data point, I have been playing with a
programmed I/O disk interface and get about 0.250MB/s. Which beats the
RQDX3...

Tim: If you have your benchmark program available and it would run on
V5.03, I would be interested.

I was just looking at the Qbus protocol diagrams and it looks like the
absolute minimum cycle time is 350ns. The maximum transfer rate is then
5.714MB/s. This is assuming continuous block mode DMA transfers. The
protocol limits it to 16 words without a bus arbitration. The true
absolute maximum is less. There is no allowance for memory speed either.

Am I correct about this?

-chuck

[sm6nnc]

Post by Charles H Dickman

Post by Tim Shoppa
I did (and published) some benchmarks of SCSI MSCP-emulating
controllers probably a decade ago. And indeed the CMD CQD 440 was the
winner. But they all beat the pants off a RQDX3!
Here are the peak data rates measured for read and write 64
Read Write
---------- ----------
Andromeda SCDC 2.298 MB/s 1.131 MB/s
CMDCQD440 2.397 MB/s 1.525 MB/s
CMD CQD220 1.418 MB/s 0.882 MB/s
CMD CQD220A 2.088 MB/s 1.409 MB/s
DEC RQZX1 1.379 MB/s 1.097 MB/s
Viking QDT 0.846 MB/s 0.704 MB/s
DEC RQDX3 0.164 MB/s 0.161 MB/s
The benchmarks were done under RT11FB 5.7 doing 1, 2, 4, 8, 16,
32, and 64 block-at-a-time READW's and WRITW's to 16384-block
data files. A KDJ11B (PDP-11/73) CPU with 2 Megabytes of Clearpoint
non-PMI memory was used for the bencharmks. With the SCSI
controllers a Barracuda 7200 RPM ST15230N drive was used; with
the RQDX3 a RD52 drive was used.
Tim.

Interesting. To add an additional data point, I have been playing with a
programmed I/O disk interface and get about 0.250MB/s. Which beats the
RQDX3...
Tim: If you have your benchmark program available and it would run on
V5.03, I would be interested.
I was just looking at the Qbus protocol diagrams and it looks like the
absolute minimum cycle time is 350ns. The maximum transfer rate is then
5.714MB/s. This is assuming continuous block mode DMA transfers. The
protocol limits it to 16 words without a bus arbitration. The true
absolute maximum is less. There is no allowance for memory speed either.
Am I correct about this?
-chuck

Interesting, indeed! I'm about placing an order, I found one at a
price that's god enough, and this benchmark sent "god wibrations" on
the type!

And regarding CMD-types: Yes; I agree to your old memories, and want
to add...
Someone wrote: The CQD 420 is a 440 without the differential side
populated
440 is surely a quad-wide card. I THINK i remember some writing of a
420 being dual, no guarantees given!

/TM means TMSCP & MSCP (ie Tape & Disc combined on the same SCSI-
bus)
/TMP means TM + "PassThru" (I GUESS this means: no internal
terminator)
/TMS means TM + "Hardware Shadowing"
TMP and TMS taken from a 440/443 manual I found on-line.

I have seen some different models, but NEVER a 420 or 420 /T or /M.
That was the older days...

I have also found:
CQD 200...
CQD 220/TM ?
CQD 220A/TM SCSI-2 5 MByte/sec peak
CQD 420/TM Fast SCSI-2 up to 10 MByte/s peak
Add 3 to any model number to get S-handles... I dont known how "old"
models this is valid for, though.

So, now I will have to wait for shimpment and se if a manual is
included...

I'm sorry, I didn't write data of those old controllers when they
passed my screen last night in search of the more modern types...


Best regards, and thanks for help, /Göran

[Tim Shoppa]

Post by sm6nnc

Post by Charles H Dickman

Post by Tim Shoppa
I did (and published) some benchmarks of SCSI MSCP-emulating
controllers probably a decade ago. And indeed the CMD CQD 440 was the
winner. But they all beat the pants off a RQDX3!
Here are the peak data rates measured for read and write 64
                  Read           Write
                  ----------     ----------
Andromeda SCDC    2.298 MB/s     1.131 MB/s
CMDCQD440       2.397 MB/s     1.525 MB/s
CMD CQD220        1.418 MB/s     0.882 MB/s
CMD CQD220A       2.088 MB/s     1.409 MB/s
DEC RQZX1         1.379 MB/s     1.097 MB/s
Viking QDT        0.846 MB/s     0.704 MB/s
DEC RQDX3         0.164 MB/s     0.161 MB/s
The benchmarks were done under RT11FB 5.7 doing 1, 2, 4, 8, 16,
32, and 64 block-at-a-time READW's and WRITW's to 16384-block
data files.  A KDJ11B (PDP-11/73) CPU with 2 Megabytes of Clearpoint
non-PMI memory was used for the bencharmks.  With the SCSI
controllers a Barracuda 7200 RPM ST15230N drive was used; with
the RQDX3 a RD52 drive was used.
Tim.

Interesting. To add an additional data point, I have been playing with a
programmed I/O disk interface and get about 0.250MB/s. Which beats the
RQDX3...
Tim: If you have your benchmark program available and it would run on
V5.03, I would be interested.
I was just looking at the Qbus protocol diagrams and it looks like the
absolute minimum cycle time is 350ns. The maximum transfer rate is then
5.714MB/s. This is assuming continuous block mode DMA transfers. The
protocol limits it to 16 words without a bus arbitration. The true
absolute maximum is less. There is no allowance for memory speed either.
Am I correct about this?
-chuck

Interesting, indeed! I'm about placing an order, I found one at a
price that's god enough, and this benchmark sent "god wibrations" on
the type!
And regarding CMD-types: Yes; I agree to your old memories, and want
to add...Someone wrote: The CQD 420 is a 440 without the differential side
populated
440 is surely a quad-wide card. I THINK i remember some writing of a
420 being dual, no guarantees given!
/TM means TMSCP & MSCP   (ie Tape & Disc combined on the same SCSI-
bus)
/TMP means TM + "PassThru"   (I GUESS this means: no internal
terminator)
/TMS means TM + "Hardware Shadowing"
TMP and TMS taken from a 440/443 manual I found on-line.

Ahah, now I remember: The "PassThru" was a way of sending SCSI
commands through to raw devices. I asked around at the time it was new
(late 80's?) and was told it was for doing CD and magtape magazine-
changing but required some custom drivers to get it done.

All of them (maybe exception of RQZX1) had selectable termination that
could be put in and out. At least one brand (maybe CMD, maybe
Andromeda?) could have terminators turned on and off by software
configuration.

Hardware shadowing makes sense where you need shadowing... I know that
some RSX versions supported hardware shadowing for MSCP. Seems like
way overkill for a hobby application.

As to the source to my benchmark program, it was literally just
some .READW's and .WRITW's, probably done through Fortran for off-the-
cuffness. Since it's RT-11, by definition the blocks are
contiguous :-).

You're probably right about the 440 vs 420. I think both were quad-
height but I'm really stretching my memory again.

Tim.

[Jerome H. Fine]

Post by Tim Shoppa
I did (and published) some benchmarks of SCSI MSCP-emulating
controllers probably a decade ago. And indeed the CMD CQD 440 was the
winner. But they all beat the pants off a RQDX3!
Here are the peak data rates measured for read and write 64
Read Write
---------- ----------
Andromeda SCDC 2.298 MB/s 1.131 MB/s
CMD CQD440 2.397 MB/s 1.525 MB/s
CMD CQD220 1.418 MB/s 0.882 MB/s
CMD CQD220A 2.088 MB/s 1.409 MB/s
DEC RQZX1 1.379 MB/s 1.097 MB/s
Viking QDT 0.846 MB/s 0.704 MB/s
DEC RQDX3 0.164 MB/s 0.161 MB/s
The benchmarks were done under RT11FB 5.7 doing 1, 2, 4, 8, 16,
32, and 64 block-at-a-time READW's and WRITW's to 16384-block
data files. A KDJ11B (PDP-11/73) CPU with 2 Megabytes of Clearpoint
non-PMI memory was used for the bencharmks. With the SCSI
controllers a Barracuda 7200 RPM ST15230N drive was used; with
the RQDX3 a RD52 drive was used.

Jerome Fine replies:

I would VERY much appreciate an opportunity to run the identical
benchmark and compare it with the ESDI controller that I have,
a Sigma RQD11-EC. Any possibility of obtaining the code?

Sincerely yours,

Jerome Fine