> > [.. .stuff about CSE ...]
> Does that actually work if the disk is online to both systems at the
> same time? I have visions of Linux not appreciating that the
> data on its
> ro volume is actually dynamic.

The point of CSE is to allow a VM userid to log in on any system that is
part of the CSE complex and have it's normal mindisks and spool files
available to it.  Note that spool sharing requires PVM, which does not
yet have a license for IFLs, even though it works fine, and there are
some significant setup implications to get CSE working in a local
context. The way it works is shown below.

Say you have 2 systems bound together in a CSE complex, NODEA and NODEB
(either two separate boxes or two LPARs, etc). So, when you log on a
Linux system in a VM userid on NODEA, the minidisks containing the data
go R/W on NODEA and R/O everywhere else. There are still the consistency
problems because the mindisk directory is still updated in core just
like always.  IF, however, you have a way of LINKing or detaching a
minidisk and signaling another system to pick it up, the "other" system
can be on some other node in the CSE complex.

CSE is useful as a beginning for HA systems and shared DASD. It's not a
complete solution just yet, but it does handle a few of the harder
parts. Building something like GFS or GPFS on top of CSE would be an
intriguing possibility, if either could be taught a few new tricks about
how to hand the disks back and forth.

-- db

Reply via email to