[email protected] (Phil Smith III) writes:
> And I'm 99.9% sure that DASD capacity was determined by building the
> geometry and then trying various densities until error rates became
> unacceptable, then backing off slightly. Which would explain the
> weird, random sizes with each generation (until 3390, after which it
> went to arrays and became standardized-on what future generations will
> consider a weird size).

error detecting/correcting started moving to fixed block sizes ... aka,
FBA ... even 3380 had fixed cell size for error correcting ... however
POK's favorite son operating system has had difficulty weening off
CKD. There hasn't been real CKD made for decades, all being simulated on
industry standard fixed block disks.

the recent move from 512byte to 4096byte fixed blocks is largely
motivated by error correct.
fixed-block
https://en.wikipedia.org/wiki/Fixed-block_architecture
FBA 512->4096 migration
https://en.wikipedia.org/wiki/Advanced_Format

original "raid" patent was by IBMer in the 70s
https://en.wikipedia.org/wiki/RAID

first use was IBM S/38 ... because common single disk failure took out
everything. part of S/38 organization simplification was scatter
allocation across all disks (treated as single filesystem) ... and
therefor any single disk failure took out the whole system ... all disks
had to be backed up as single unit/pool (because of scatter allocation)
... and any recovery required complete system restore.

Note originally 3380 had 20 track spacings between each data-track ...
flying lower met adjacent tracks had less interferance and cut the data
track spacing in half (double-density with twice the number of
tracks/cylinders), then spacing cut again for triple-density (three
times the number of tracks/cylinders).

trivia: I got dragged into idea the IBM "father of risc" had for
"wide-head" 16 adjacent datatracks with servo tracks on either side
... read/write all 16 simultaneously (while tracking servo tracks on
both sides of the data tracks). This was in 3090 and 3380 triple-density
time-frame. The problem was that the IBM mainframe data transfer is
16*3mbytes/sec or 48mbytes/sec. Even when ESCON is announced in 1990
(with ES/9000, when ESCON is alread obsolete) it is only 17mbytes/sec.

little more trivia: 70s, engineer was running "air bearing" (floating
heads) simulation (part of reducing head flying height enabling greater
densities) on research 370/195 ... but only getting a couple turn
arounds a month (even with priority designation). I had done enhanced
bullet proof, never fail operating system for bldg 14&15 allowing them
moving from stand-alone testing to doing concurrent development testing
under operating system. Turns out even concurrent testing only used
percent or two of processor ... so we set up private online service
using the machines. Bldg15 had 2ndor3rd engineering 3033 from POK, and
we get the air bearing simulation moved over to the 3033, where he can
get several turn-arounds a day (even tho 3033 has little less than 1/2
processing of 370/195).

more trivia: I had done channel-extender support in 1980 for STL that
was moving 300 people from IMS group to offsite bldg ... but the POK
people playing with what becomes ESCON ... blocks it release to
customer. In 1988, I'm asked to help LLNL standardize some serial stuff
they are playing with which quickly becomes fiber-channel standard,
including some stuff that I had one in 1980 (FCS, originally
100mbyte/sec concurrent in both directions).

Then some POK engineers get involved in FCS and define a protocol
that radically cuts the native throughput ... which is eventually
released as FICON. Most recent published FICON numbers I've seen
is peak I/O z196 test that used 104 FICON (running over 104 FCS)
getting 2M IOPS. About the same time there was FCS announced
for E5-2600 blade claiming over million IOPS (two such FCS,
getting more throughput than 104 FICON running over 104 FCS).

-- 
virtualization experience starting Jan1968, online at home since Mar1970

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to