🚀 go-pugleaf

In article <k2.721304278@woodstock>, k2@bl.physik.tu-muenchen.de (Klaus Steinberger) writes:
> tgcpwd@rw9.urc.tue.nl (Wim van Dorst) writes:
...
> I did have VMS Posix running on my VAX, and it gives indeed better
> results. I'm seeing results in the 200 - 300 kBytes/sec range.
> (A little MVIII, with SMD disks, platter rate around 2Mb/sec).
> I did not know how well the Posix environment did emulate Unix
> system calls, so I think native VMS programs with their ugly $QIO
> interface will see better rates.
> 
Yep, but you will need to specify contiguous files, lock the buffers into your
working set, run at a high real-time priority... Them you should get xfer rates
that aproach the platter speed of the drive.
On a Q-bus RSX ( PDP-11 ) system years ago I could pull >2MB/sec off 3 Fugitsu
ESDI drives. This was mainly due to defered writes in the controller cache.
The RSX case was multiple tasks BTW, not a single read stream. The best IO with-
out a cache is ~1.1MB/sec from 1.2MB/sec SMD drives, dual ported to an 11/70.
Again this was multiple streams.

A VMS system will also show how bad fragmentation can be to performance. Just
remove the contig bit from the create and turn it lose on a near full busy
disk!

All this only proves that we are infact measuring everything EXCEPT disk speed!!

~Paul