Shmuel Metz , Seymour J. wrote:
Not quite; the 370/168 had two type of block multiplexor channel,
single byte and 2 byte. The 2 byte channel ran at 3 MB/s. The 3880
only supported the single byte channel, which was less expensive.
re:
http://www.garlic.com/~lynn/2007e.html#40 FBA rant
about the only thing that i remember that would use the 2byte/3mbyte/sec
channel was the 2505-1 fixed head disk.
http://www-03.ibm.com/ibm/history/exhibits/storage/storage_2305.html
2505-2 was 1.5mbyte/sec, 11.2mbyte capacity, 5millsecond avg. rotational delay
2505-1 was 3mbyte/sec and 5.4mbyte capacity ... heads were positioned i.e. so
the
avg. rotational was cut in half from 5milliseconds to 2.5millseconds, data
transfer
rate doubled from 1.5mbytes/sec to 3mbyte/sec, and data capacity cut in half
from
11.2mbyte to 5.4mbyte.
I never actually knew of customer with either 2305-1 and/or 3mbyte/sec (2byte)
channel
Nominally bus&tag cable channel was rated at 200ft max. distance ... but even 2505-2 (at
1.5mbyte/sec) could present problems. We had 370/158 that had troubles with (2305-2 controller)
2835 down around 50ft that was fixed by configuration with shorter cable.
Data streaming (reduce the requirement end-to-end synch on every byte) allowed
for both
higher data rate as well as doubling maximum channel cable distances.
Slightly related from long ago and far away ... effectively "native mode"
for electronic paging disk is very close to FBA kind of operation (w/o
the rotational delay)
From: wheeler
Date: 08/05/82 16:17:32
re: 1655 electronic disks; Native mode operation has the same
performance as 2305 simulation ... not faster, no slower.
However, in native mode all 12meg worth of drum is used as data
blocks. In 2305 simulation mode, only that amount of formated space is
used for data blocks. VM uses a format which only utilizes approx.
9.5meg worth of data blocks (the rest is inter-record gaps and dummy
block spacers to optimize slot sorting). The result is native mode
represents about a 30% increase in drum space (an 1655 box with 4
simulated 2305s becomes the equivalent of 5.3 2305s in native mode).
They have been saying they would have a 3meg. data streaming option
available by August for 1655. That would mean twice the data transfer
rate compared to either a real 2305 or an 1655 simulated 2305. I
haven't confirmed it, but it was my understanding that 3meg. data
streaming would be available for either 2305 or native mode.
SJRLVM1, SJEVM5, and at least one machine in STL are running 1655s (48
meg./4 drum) in 2305 mode. They are all 1.5meg. versions. In addition,
SJRLVM1 has a data streaming STC 2-drum electronic device (3
megabytes) ... & the STC drums don't have a native mode option. We
also have a combination of real 2305s and 3380s and are in the process
of running various performance comparisons.
Note: at 1.5meg. mode, an electronic drum has the same maximum
thru-put capacity as a 2305 drum ... under VM at maximum load, there
are long CCW chains transfering multiple page requests in one SIO
operation. The data transfer is the same, so the electronic drums
don't buy anything there. It is in the area of average access time
that electronic drums improve performance. A 2305 drum has a 5
milliscond avg. rotational delay (access delay) per SIO. An electronic
drum has avg access delay of 300-400 microseconds (approx. 1/20th of a
2305). Time to transfer one page is approx. 2.7 milliseconds for
either devices at 1.5meg. transfer. For long CCWS chains with one
rotational delay per 20-30 pages transfered performance is about the
same:
chain 2305 [EMAIL PROTECTED] [EMAIL PROTECTED]
size elapsed elapsed elapsed
1 page 7.7mills 3.0mills 1.6mills
2 page 10.4mills 5.7mills 2.9mills
5 page 18.5mills 13.9mills 7 mills
10 page 32 mills 27.4mills 13.9mills
20 page 58 mills 54.4mills 27.4mills
On a moderately loaded, page bound system, electronic drums can
significantly improve the paging performance.
... snip ...
above comment about moderately loaded, page bound system ... i.e. w/o
long page CCW chains so real rotational delay was significant part of
service time per page transfer.
As part of my resource manager shipped in the mid-70s ... I had also
shipped page "migration" support (pages that became inactive on
"higher speed" devices would be migrated to "lower speed" devices)
... improving effectiveness of higher speed devices for paging
operation.
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
