Re: Bufferpools/hiperpools & memory

Joel Goldstein

Re: Bufferpools/hiperpools & memory
(in response to Isaac Yassin)
Isaac,
If your system is really paging to disk on a regular basis (not sure what
your .5 number is... per/second?)... you need to look at your paging
rate to expanded... which is probably much too high to have good
performance.
High MVS paging rates to expanded make performance worse, even when
the I/O wait component drops, and the BP hit ratio goes up.
You need to add all the page-ins across all your pools, and then
calc a rate/sec.
Also calc an MVS overhead factor for your transactions.
(C2 Elapsed-C3 Wait)/C2 CPU. If this > 1.5 you are constrained, and
part of that constraint may be paging (especially if your processor is
< 95% busy.
Regards,
Joel


Message text written by DB2 Data Base Discussion List
>I did an interesting "exercise" during the last weeks.
VBPs were enlarge nearly up to the max allowed - 398500 pages all together.
System paging to disk averaged 0.5 on the peak time as measured by SMF.
No problems were encountered due to being near the edge.

Hiperpools for BP7 (DSNDB07) - I tended not to allocate HBP for DSNDB07 but
during the last year
I did many test with it and found out you may benefit even if sometimes it
looks like a waste (which it is not).
HBP is used by DSNDB07 when needed - so if your sort can finish in the BP -
then fine, otherwise you
get to use the HBP for reading the pages for the continuation of the sort
and save I/O to the disk to read
those pages. Afte heavy sorts you can see that your HBP is really backed up
by expanded storage.
Afte some time of not using it (smaller sorts) then those pages will be
given to other expanded storage demands. BUT you need to use CASTOUT=YES
for
that.

I had 3.7 GB real memory on the machine - 2 GB for central storage and 0.7
for expanded storage.
The HBP allocations were 1 GB together. Using CASTOUT=YES gave the option
to
"move" allocations of HBP as needed by DB2 and the MVS as well.<



Max Scarpa

Re: Bufferpools/hiperpools & memory
(in response to Joel Goldstein)
Hi Isaac, what's is your DB2 (MVS) UIC ?

Regards

Max Scarpa



Isaac Yassin

Bufferpools/hiperpools & memory
Hi,

I did an interesting "exercise" during the last weeks.
VBPs were enlarge nearly up to the max allowed - 398500 pages all together.
System paging to disk averaged 0.5 on the peak time as measured by SMF.
No problems were encountered due to being near the edge.

Hiperpools for BP7 (DSNDB07) - I tended not to allocate HBP for DSNDB07 but
during the last year
I did many test with it and found out you may benefit even if sometimes it
looks like a waste (which it is not).
HBP is used by DSNDB07 when needed - so if your sort can finish in the BP -
then fine, otherwise you
get to use the HBP for reading the pages for the continuation of the sort
and save I/O to the disk to read
those pages. Afte heavy sorts you can see that your HBP is really backed up
by expanded storage.
Afte some time of not using it (smaller sorts) then those pages will be
given to other expanded storage demands. BUT you need to use CASTOUT=YES for
that.

I had 3.7 GB real memory on the machine - 2 GB for central storage and 0.7
for expanded storage.
The HBP allocations were 1 GB together. Using CASTOUT=YES gave the option to
"move" allocations of HBP as needed by DB2 and the MVS as well.


Isaac Yassin
DBMS & IT Consultant
[login to unmask email]



Isaac Yassin

Bufferpools/hiperpools & memory
(in response to Max Scarpa)
Hi,
Sorry the reply botton decided to go on vacation.
Max - UIC 254, exp-sto migration age is 3,600+ at peak time(seconds)
otherwise it's 150,000+ means pages reside there for enough time.
Joel - system pages to disk on the peak time at 0.5 page/second (forgot to
put it in my prev note) but only on very busy times, otherwise it's flat
zero. Page rate to expanded at peak time fluctates between 0 - 200 per
second (according to tmon/mvs). CPU goes 100% at the peak time - all the
time. Page-in at VBP (all together) is less then 20,000 . Workload combines
both ad-hoc and long scans.
I started enlarging VBP together with the system group to see how it affects
the machine after we added more storage. They don't complain up to now...


Isaac Yassin
DBMS & IT Consultant
[login to unmask email]