🚀 go-pugleaf

ha...@watson.ibm.com (hack) writes:
> Historically, the coupling between virtual card punches and readers (across
> the RSCS company-wide network, aka VNET) was the primary communication
> interface (nearly 30 years old).  It may sound quaint, but it's really just
> a data pipe (with 80-byte blocks) with tagged ends to perform the connection,
> and it has a really simple interface (no connection protocols etc. -- the
> network is store-and-forward, so it's literally like mailing postcards where
> bunches (individual "spool files") are guaranteed to arrive in order (there
> is no virtual equivalent to a deck of cards spilled on the floor).

one of the problems with the "spool file" interface was that it was
syncronous on open, close and on 4k buffer boundaries doing various
disk I/O (whether it was using straight UR simulation or 4k block
transfers). in the mid-80s, a VNET virtual machine supporting even a
few full-duplex 56kbit links could spend 90% of elapsed time in
"blocked" state waiting on syncronous (disk) spool file operations.

one of the things that we did in HSDT was implement several
asyncronous functions for VNET so we could start seeing it drive
multiple full-duplex T1 links at media thruput.

In HSDT, we also got the mainframe TCP/IP thruput up from 44kbybtes/sec
to hardware channel speed.

random refs:
http://www.garlic.com/~lynn/subtopic.html#hsdt
http://www.garlic.com/~lynn/subtopic.html#xtphsp
http://www.garlic.com/~lynn/internet.htm

-- 
Anne & Lynn Wheeler   | ly...@garlic.com -  http://www.garlic.com/~lynn/