If we're talking about utilizing multiple channel paths to devices,
multipath I/O has been around for ages in the 390 world -- at least 10 years
that I can remember, and even earlier with the primary/alternate control
unit support (max 2 paths for that far back).  It really got practical with
the XA I/O architecture when all the gory detail information about paths got
moved down into the IOCP layer and the OS no longer had to deal with it
directly.

There are several layers in the 390 I/O subsystem:

devices
control units
channel paths
device numbers

The relationship of the first 3 items is handled by the IOCP definitions in
the processor element. IOCP describes what devices are connected to what
control units, and what channel paths are connected to each control unit.
Those combinations are assigned unique device numbers, which is the only
thing an operating system sees as devices that it can do I/Os against --
IOCP hides all the details of how the cabling is set up and managed. So, by
virtue of using the XA I/O subsystem on the 390, Linux gets multipath I/O
for free -- it's "just there".

(BTW, a ESS with 16 channel paths is a *very* fun toy. Linux works very
nicely...8-))

-- db

David Boyes
Sine Nomine Associates

Reply via email to