This is for the rest of you, Damon is right across the hall and can
get this info quicker than the long way via across the country and
back.
#1 Currently ADSM is good enough to beat both you and your hardware
into the ground! That is you will definitely discover what your weak
links are in your computing environment.
#2 'Cause of #1, MVS has discovered that 3390 Mod-9's are SLOW ! ! !
(that was already known but it is REAL CLEAR NOW )
#3 MVS & ADSM will only take disk pool storage files to the limit of
32-bit addressing THUS a dedicated mod-9 to a diskpool requires
multiple files to reside on it in order to fill it.
#4 MVS was feeling left out due to all the AIX ADSM environments so to
make it feel better it now has a 100+GB diskpool and enough clients to
make it beg for forgiveness...
#5 ADSM, by observation, has some real good algorithm for spreading
out the work/data across available resources AS FAR AS FILES/VOLUMES
DEFINED TO ADSM (it doesn't break it down to op sys levels, ie
physical devices, just logical items)
#6 If you have 6 files/vols per physical device, 10 devices per
diskpool... random chance dictates that sooner or later you will have
only 6 sessions and they will all be going to the same physical device
(sigh) Oh, and increased workload triggers increased frequency of
random events !
7# I'll answer the question below now....
Spread EVERYTHING out as much as you can ! across : physical dasd,
across controllers, across channel paths, etc...
Especially the adsm data base...
(now I've done this under AIX with SSA disks but I'll describe and
link to MVS terms...)
my MOST spread out environment looks like:
4 SSA cards (aprox mvs CHP), ssa cards are in pairs to cover
failure/roll over (multiple CHP's), each pair of SSA cards (acting as
a single card) has an A-loop (mvs dasd controll unit) and a B-loop
(mvs controll unit), each loop has 16 drives (mvs string of dasd
behind a controller)
Now actually a "ssa loop" is an A1 port & an A2 port where the first
drive off A1 is the last drive off of A2, A1 & A2 act independently
but if A1 fails the first drive off it can still be reached as the
last drive off A2, just a longer path... SOOOooooooo
So I've got one db file on first disk off of card(s) (1/2) A1 & A2,
these db vols are mirrored off of card(s) (3/4) A1 & A2; the log is
first vol off of card(s) (1/2) B1, and the mirror is the first vol off
of card(s) (3/4) B1
Compared to MVS that would be each db vol is on the head of string of
dasd, each on different CHP's, the primary DB vol's control units
would be located on each other's secondary CHP, the copy DB vol's
control units would be on TOTALLY different CHP's and each of their
control units would be on each other's alternate CHP....
Then the log volumes would be on the head of string of the second
string of dasd behind the dbvols which would also give them alternate
CHP's
CAN YOU SAY : S P R E A D O U T T H E W O R K L O A D !
and a lot of stuff's gonna hav'ta die before ADSM goes down...
really sad is the clients are still killing this box !
I'm starting to lean towards solid state dasd for the DB & LOG.
later
Dwight
______________________________ Reply Separator _________________________________
Subject: Re: Diskpool configurations on MVS.
Author: ADSM-L (ADSM-L AT VM.MARIST DOT EDU) at unix,mime
Date: 5/21/97 11:27 AM
Damon,
I hope you get some responses on your question. Currently our bottleneck
is our network, but we are attempting to implement our MVS server in as
efficient a manner as possible to prevent other bottlenecks when our
network is upgraded.
My main question is whether it is better to have a dasd pool comprised of
many small chunks. This seems the logical conclusion, but I would
appreciate some confirmation.
Good luck,
Dan T.
----------
> From: Damon Taylor <dtaylor AT AMOCO DOT COM>
> From: Damon Taylor <dtaylor AT AMOCO DOT COM>
> To: ADSM-L AT VM.MARIST DOT EDU
> Subject: Diskpool configurations on MVS.
> Date: Wednesday, May 21, 1997 9:20 AM
>
--MimeMultipartBoundary--
=========================================================================
|
