From: Glenn C. Everhart
To: Pat St. Laurent
Date: 24 March 1997

Status:

MultiPath Failover:

Practically all of the time in March has been devoted to multipath 
failover work. A review was held resulting in some edits to the design
specification, and in which I got answers to several questions I had
wanted to get answers for at the review. The updated document is now
in star:: in docd$:[evms.project_documents]ds-multipath-switch.ps.

I have now done initial coding of the production switching driver
and am cleaning up its build. This is directed at producing working
switch code as quickly as possible to facilitate testing as the remaining
pieces of the project are coded. This driver is written in C (with a tiny
Macro-32 stub to deal with some routines which cannot be handled by the
C compiler, which is also written) and is close to the projected length
of 2500 lines (actual is 2383).

This is still within the schedule laid out in the project plan, or perhaps
slightly ahead.

QARs
I worked on another problem with DKdriver IO$_AUDIO this month and have
written code which fixes the identified problems. It is being reviewed
as I write this. Since it checks ALL buffers for legality now which are
passed by IO$_AUDIO, I believe it will eliminate crashes caused by
invalid I/O requests of this sort.

I also worked on a PKZdriver problem Bill Clogher was having and he
reports that a fix I gave him prevented the system hang he saw. I expect
to close that QAR today or tomorrow (unless his later tests find some
additional problems).

Other Issues

I have also found a stack limit check routine, and written code to deal with
a long standing problem with REQCOM (which multipath needs to have fixed)
using it. Initial tests with IOSUBNPAG and DKDRIVER calling this routine
show it to be functioning as expected and causing no problems.

I discussed IRP changes I have proposed in the multipath failover DS
with Chris Whitaker of EDO. It seems that EDO also needs such features,
for the snap capable disk project they are working on. Lenny Szubowicz
and I have had some discussions about this, and need to schedule more;
I will write up a more formal proposal for the interface in this
connection, based on discussions at the design review (at which time the
EDO needs were not known).

Per a suggestion by Steve Z., I passed a letter describing some ideas I
had for generalizing the notion of file directory to Bob Supnik so that
if the labs are interested in them they'll be able to look them over.


Glenn Everhart