Hopefully Jim Mulder or Pete Relson can jump in if they disagree.

As far as I can determine this recommendation hasn't changed much since ESA/390.

I use ( REAL STOR FRAMES *3 ) for total page slots. This is per LPAR.

Allocate total local paging space as  (total page slots/0.35) to facilitate the 
block paging slot allocation algorithm.
I recommend a max of 100 pps/device. (personally, I dedicate the device to 
paging).

Spread over as many devices/paths as practical.


The "optimum" VSTOR config is 2:1 virtual to real. I personally use 3:1 (RSTOR 
*3). This is a trade off between page rate and devices allocated.

I.e. the sum of all working sets for the LPAR(in frames)  is up to 3 times the 
number of real storage frames available.
This paging rate goes up *VERY VERY* quickly after 3:1 IIRC, this is a SQUARE 
function.
For a given original page rate:
To go from 2:1 virtual/real to 3:1 virtual/real generates the ratio (9/4 * 
original page rate) 2.25 times
To go from 2:1 virtual/real to 4:1 virtual/real generates the ratio (16/4 * 
original page rate) 4 times the original page rate.

Additional comments interspersed below.
HTH,

<snip>
We are implementing a DS8800 with Hyperpav and like to give ASM the best paging 
configuration. However, the recommendations in Initialization and Tunig Guide 
are outdated, speaking of 3380 devices, Seek penalties, dedicated devices, 
channels and control units and the Suspend/Resume logic. Occasionally PAV is 
mentioned, as well as z/OS 1.8, where the Suspend/Resume is no longer used, 
without explaining the consequences.

 

There is no good recommendation how to set up a good paging configuration on a  
modern DS8800 device, nor does it explain the consequences of Hyperpav and 
WLMPAV set to NO in relation with OA32453 where ASM is still building a " I/O 
infrastructure" that utilizes Hyperpav. Nowhere is explained what this 
infrastructure looks like. From conversations with Greg Dyke when he was the 
expert in that area, I know that in the WLM managed PAV environment ASM 
reserved 2 access paths to each page dataset, resulting in 2n-1 aliases 
dedicated to a paging volume.


 Is this still done so in the Hyperpav situation?

 >>>>>As far as I know, this is still the case

Our production Sysplex has 10 LCUs on the DS8800 and is connected with 8 Ficon 
channels. Hyperpav can select alias devices within an LCU, not over LCUs.

I can allocate 10 page volumes, 1 per LCU, with 1 local page dataset per 
volume. This would give ASM 20 access paths to its paging configuration.


I can allocate 2 page datasets on each volume, giving ASM 40 access paths to 
its paging configuration. Or 60 or 80.

However, this all has to be handled by the 8 Ficon channels.

 

How do I determine the optimum?
>>>>> See above

 

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to