🚀 go-pugleaf

I don't know whether CP's screwing with channel programs prevents
this, but there is no reason why (with the right O.S., say, MVS), you
can't read every block on a cylinder in 15 revolutions.  (In fact,
using QSAM with lots of buffers can do nearly as well as EXCP using
obnoxiously unfriendly-to-other-jobs channel programs.)  Head
switching is done at electronic speed -- the arm is not moved.  The
simplest channel program to do this would be 15 {seek, read r0 (to
orient), read-multiple-ckd}.  Of course, you need a 725K buffer to do
this, and as someone else already mentioned, you need to follow the
count fields through the buffer to locate the individual records.  A
3380 really can transfer data at the documented speed of 3MB/sec.
Actually, it's 47476*60 = 2,848,560 bytes/sec.  The CMS "filesystem"
constrains this to 4K*10*60, or less than 2.5MB/s.  Note that 4K
blocks on a 3380 waste 14% of the 3380 track capacity.  Now... under
CMS, how can you even insure that the blocks in a file are in
sequential locations on the disk?  I thought CMS splattered them
wherever it finds a free block on the mdisk...  This is the unix/
tops20/MSDOG/[everything-but-MVS] concept of "let's be user friendly,
performance be damned" philosophy.

/Leonard

P.S.  Even though I've thrown out the bait, we don't really need
another "OS wars" discussion on this list...