Well the time has come to gather up my views of Wednesday. I spent 8:00
to 10:00 revelling in Alfredo's "Always On (topic)" session. And Alfredo,
despite all your stated concern, I think you have little to worry about. There
are always last-minute adjustments to the topics presented at the conference
and we don't mind if you make those adjustments continuously throughout the
session :-). Besides, Alfredo had a (hidden) agenda (which he showed us
briefly).
Then I skipped out on Ann Livermore and did some programming, so I missed the
amazing Ann-mentioned-the-3000 incident. Perhaps someone who was there could
report in more detail. 11:30 saw the start of SIGSysMan. We heard from
Walter McCullough on some cool new stuff (I think that's a technical term).
I don't spend much time in the HA world, but what I gathered was that we can
no hook the HP to fiberchannel and to SAN fabric (someone correct me if I have
something wrong here). After that, we heard from Jon Diercks, there to give us
the latest word about the _MPE/iX_System_Administration_Handbook_
(see http://diercks.net/mpe/ for more information). Looks like it should be
well worth getting. Jon says that the intended audience includes everyone from
those new to the box to experienced MPE admins.
Jeff Vance arrived and rather than let him eat, we made him get up and talk. He
reported on the many SIG items which HP has finished and a number which are
either under way or planned. Jeff, I don't trust myself to get the list right,
could you give it us a quick rundown here?
Donna then thanked HP for the many things they've accomplished for us--in
particular for the "enhanced" ping which Lars Appel put on Jazz, the improved
Patch/iX and that the instructions which come with a patch tape now included
Patch/iX as an installation method.
We then went through the various things which are candidates for the next SIB.
System CI variables came up quite a bit--both there and in SIGMPE. Just in
case anyone has missed it, there is an alternative option. There is a set of
UDCs on Jazz which very nearly implement all that this item is asking. Jeff
said that using the file system to implement them is essentially what CSY would
do, so if you're interested in global CI vars, give the UDCs a try.
There was a quick report on the System Software Update manual. While there is
concern that the process needs improvement, there is one point of immediate
trouble and that is that the Instant Information version of the manual which
came with 6.5 has numbering problems--II evidently decided to do its own
numbering, so the section numbers you are sent to don't match II's numbers for
the section. If you need the manual, use the pdf version instead. Likewise
for the version on docs.hp.com--go with the pdf version.
I missed out on some of SIGMPE and don't have a copy of the 28 items we
discussed there, so I'll leave that report for someone else to give. Just a
quick note regarding space--I don't know if much could have been done about
space, but both of those SIGs were pretty crowded, as was Kevin Cooper's talk
on MPE/iX Performace Updates. Regarding the latter, I only got in at the
tail end as it was across the second hour of Alfredo's talk.
From the part I saw, there are two patches which may improve performance on
6.5 or 7.0. MPELXH8 provides a "make_absent" option (call?) so that priv-mode
processes can explicitly free memory pages they no longer need, saving the
trouble of Memory Manager having to figure that out itself. It also reduces
how often Memory Manager sets out to make free pages and how many it tries to
make free at a time. Evidently on very large-memory systems, Memory Manager
was getting a bit carried away. It also reduces the overhead of mapping closed
files out of memory and tells Memory Manager to relax a bit in flushing dirty
pages when the I/O system is busy. The example he gave was that everyone shows
up to work in the morning and allocates a whole bunch of memory, much of which
they don't use (loaded libraries, seldom-used parts of code, stack) and then,
when it comes time to flush those on the big system, Memory Manager just
plugged right ahead, doing its thing even if the I/O subsystem was already
swamped.
If you've got a big system with lots of memory, MPELXH8 is probably a good
idea. MPELXH3 is the TurboStore performance patch and makes use of the new
"make_absent" option to immediately free up all the pages of memory its done
with and no longer needs.
After SIGMPE, I headed upstairs to Mark Bixby's "Open Source Development on
MPE", which was basically Porting 101. He talked through the process, gave
advice on how to make it go smoothly and recommended 3000-L for questions
(which recommendation one hears in many of the sessions). After that, I got
down to the Expo floor for a bit. Picked up a DLT car, two magnetic stuffed
platypi (hey, when you have two kids, you have to think about these things),
a Mickey Mouse, a stuffed cheetah, a triangular highlighter (three colors) and
a keychain surfboard.
The platypi came from Platypus Technology. They have a very cool outfit there.
It has dual diskdrive/memory/battery setup. The unit functions as a disk
drive, but the actual storage is in solid-state, so data travels very quickly.
In the case of a power failure, the batteries keep the unit running long enough
to write all the memory to disk. Slick setup. No MPE driver yet, but I told
them that they should consider it. After all, the 3000 folks believe in
processing data!
Anywho. I need to try for a bit of sleep before morning is upon us, so I'll
close for now.
Ted
--
Ted Ashton ([log in to unmask]), Info Sys, Southern Adventist University
==========================================================
Mathematics is the only instructional material that can be presented in an
entirely undogmatic way.
-- Dehn, Max
==========================================================
Deep thought to be found at http://www.southern.edu/~ashted
* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *
|
