🚀 go-pugleaf

>
>In article <9205251509.AA01050@nettlerash.berkeley.edu> IBM 370 Assembly
> Programming Discussion List <ASM370%OHSTVMA.BITNET@cmsa.Berkeley.EDU> writes:
>>202. I must first provide you for some initial data:
>>
>>  My 3380 has a maximum of 885 cylinders. One cylinder consists of 15 heads
>>reading ten 4KB blocks ( + 1 record 0 of 8 bytes + a home record ) per track.
>>This device rotates at 60 rpm and the transfer rate is 3MB/sec.
>>
>>/Regis BOSSUT
>
>Sorry, but the rotational speed of a 3380 is 3600 rpm, not 60 rpm.  That's
>why your calculations are skewed.
>
>Tim Tierney
Sorry, I really meant 60 rps and not rpm.

Anyway, let me give you more info from my investigations. By building CCW
programs, I effectively reached the limit of 15 track reads in 15 disk revs.
BUT, the real flow is 2000 kB/sec. In fact, CP uses * A LOT * of CPU time to
translate guest to real CCWs and other stuff, so I can't reach the theoretical
limit of 2400 kB/sec.

My last opinion is that the DIAG A4 mechanism is the most efficient way to
achieve high disk transfer rate for the minimum of effort, if the block numbers
are arranged in such a way that the shift between 2 tracks is 2 blocks.
DIAG A4 translates directly block numbers to real CCWs, bypassing guest-to-real
CCW translations.

I think that it should be easy to write a file disk space allocator that tries
to allocate the file's block numbers to follow the rule above. Once the space
allocated, the file is accessed by the conventional routines, as FSREAD/WRITE
use the DIAG A4 mechanism (if present, of course...)

Any comment ?

/Regis