I have paging size 2048M showed from topas and 10240M showed from "lsps -a", can anyone tell what is the difference? and how to change the PAGING SIZE (showed in topas) to 8192M?
Try to use nmon command which is very usefull.. to see status of paging and memory issue nmon command and than letter m output will be somethink like this
Physical PageSpace | pages/sec In Out | FileSystemCache
% Used 97.7% 1.2% | to Paging Space 0.0 0.0 | (numperm) 31.6%
% Free 2.3% 98.8% | to File System 0.0 0.0 | Process 30.0%
MB Used 1250.0MB 7.2MB | Page Scans 0.0 | System 36.0%
MB Free 30.0MB 568.8MB | Page Cycles 0.0 | Free 2.3%
Total(MB) 1280.0MB 576.0MB
When I first read the post I was expecting to see more than one paging space listed in your "lsps -a" output from your orginal post and that only one paging space of 2048MB would be active but that is not the case. So I checked out 4 of my AIX 5 servers and two reported the same for topas and "lsps -a" but the other two report values that were not even close to each other. All 4 running the same version of AIX.
I didn't help you answer the question but now you know that your not alone.
Hi, paging space should most certainly be mirrored. Just imagine how beautifully your system will nosedive when there is a problem with paging space. IMO mirroring paging space will not impact performance very much and may even improve performance in some cases.
You are right. The paging space contains program code that does not fit into RAM. Imagine what would happen if the single disk that contains the paging space fails. The effect would be as if you lose a memory chip. Therefore you definitely mirror your paging space if you mirror your rootvg already.
I do not know what your AIX background is but as you answer questions about AIX memory management you must be experienced. However, I think here you took the wrong turn.
If you use dedicated SCSI disks for you rootvg those disks usually deliver about 150 to 300 IOPS. With those the disks turn to 100% busy and from then on your GHz Power CPUs (with currently around 4.000.000.000 cycles per second) sit there idle. Improve performance? Well... probably in a micropartion environment you manage to completely deactivate the CPU cycles for the server paging and transfering the cycles to other LPAR - these might then indeed improve their performance.
If you use SAN disks for the rootvg you may get a higher number of IOPS but never get close to anything taking place in RAM. More likely using the SAN that way will interfere with other servers that use the same physical disks in the underlying SAN boxes. So again the impact on performance is huge.
Therefore you do not want the paging space being used. Mind that I do not say that there must be no paging space. But paging space is meant as a last resort in case an unexpected request for more memory appears. An AIX administrator will always strieve for calculating how much memory the server's applications need so that it runs in RAM and make that amount of RAM available. An AIX server that uses paging space all the time is not sized correctly.
from a performance point of view, paging space should be spread across as many disks as possible. Unfortunately it is as well true, that the OS needs paging space to work properly - and if a paging space disk fails, the system will halt, on reboot the paging space will be disabled. So the best solution is to create more than one paging space - each one on a separate disk - keep them mirrored or proteced by a raid5 or vio-paired storage. Each paging area should be same size - and AIX will use the least busy one at any given moment.
Absolutely true and - sadly - very often neglected!
The reason is that AIX is trying to use available paging space parts in a "round-robin" schema per default (and, to my knowledge, this default behaviour is not changeable).
If the system has uniformly sized paging space chunks available then the round-robin usage will result in some sort-of stripeset (with stripes the size of LPs), while unevenly sized paging spaces will measurably slow down the allocation/deallocation process.
To take heed of this was more important when "disk" was a physical device and employing all the disks available improved paging performance (if only on a low level). Today the possible gain from having multiple paging spaces is minimal most times, because "disk" is many times a SAN share through a VIO-Server and spreading the paging load across several of these shares will gain not much at all.
For the reasons given above (and in case the conditions are met) it is therefore in most cases better to have only one paging space instead of several ones nowadays - and to not having to use it.
