paging space

[Midrange]

hi all,

have another query..

i observed the paging space is not balance. Is it normal and is it possible to balance it even they are residing on a different volume group?

paging01 hdisk4 appsvg 2048MB 5 yes yes lv
paging00 hdisk2 indexvg 4992MB 93 yes yes lv
hd6 hdisk1 rootvg 5440MB 4 yes yes lv

thanks.

 

[plamb]

If I remember correctly paging space usage is round-robin and it is stronely sugested to be all the same size

 

[RodKnowlton]

In pure size (not percentage), paging00 and paging01 should have equal amounts in use, with hd6 having slightly more, since it's the only paging space available during boot.

Your /etc/environment doesn't happen to have PSALLOC=early in it, does it? I've never used the early allocation policy, but I've also never seen an lsps so out of whack. Maybe early allocation works in a less round-robin fashion.



Rod Knowlton
IBM Certified Advanced Technical Expert pSeries and AIX 5L
CompTIA Linux+
CompTIA Security+

 

[kHz]

Yes they can be balanced (increased) on the fly. Not sitting in front of an AIX server now, but isn't the 93 column the percentage of that paging lv in use? If so, why is paging00 so heavily used and the others are relatively unused?

 

[hirschaj]

Equally sized paging spaces used to be important before AIX 5 but it is not a big deal anymore (they changed the algorithms). You should see no ill effects from your current paging space setup as long as you are running 5.1 or higher.


Jim Hirschauer

http://www.aixexpert.com

 

[RodKnowlton]

Jim,

Ignoring the size of Midrange's paging volumes, the troubling figure, as khz points out, is the 93% usage of paging00.

Even in 5L, I still see equal space used (measured in megabytes) on all active paging spaces but hd6, which has slightly more as I mentioned above.

Have you seen documentation that paging space is no longer allocated in a round-robin fashion with AIX 5L? If so, please share the source.

Oh, and congratulations on the CATE.

Rod Knowlton
IBM Certified Advanced Technical Expert pSeries and AIX 5L
CompTIA Linux+
CompTIA Security+

 

[hirschaj]

Rod,

I agree with you in the fact that the amount of space (actual not percent) should be roughly the same due to a round robin policy. And I have never read that the policy has changed, just that the algorithm is perfectly capable of handling unevenly sized paging spaces now. I wish I could remember where I read that, but I came across it while looking for information on another issue (as is usually the case). All I remember is that it was directly from the ibm.com website.

Now to the bigger question of why there would be such an imbalance in actual usage of paging spaces. One reason that comes to mind is the ability to deactivate paging space on the fly. Deactivating all but paging00, running for a while, then reactivating the other paging spaces could create the output shown in the original post. I'm not saying it is normal, just possible.

Oh yeah, congrats to you too on the CATE (I wonder how many people actually have that one???).


Jim Hirschauer

http://www.aixexpert.com

 

[RodKnowlton]

Deactivating all but paging00, running for a while, then reactivating the other paging spaces could create the output shown in the original post. I'm not saying it is normal, just possible.

True, it's possible, but I'm not sure how you would go about "accidently" causing such a situation. Also, for it to stay in that imbalance for any length of time you'd have to have ~ 4.5GB of process memory that gets paged out and doesn't wake back up. With the default settings, the chance of this seem vanishingly small. That's why I'm suspecting early allocation, although I don't have any experience with it. If a process were to allocate 4.5GB of memory on a machine that was already paging, early allocation would allocate it out of paging space immediately. I just can't find any mention of whether this space is allocated round-robin or not.

Midrange,

Was PSALLOC set?

Oh yeah, congrats to you too on the CATE (I wonder how many people actually have that one???).

Shortly after I got mine in September of '03, there were just shy of 800 world wide, and 192 in North America.

You can email certify@us.ibm.com for the current numbers. If you do, please post them.

Rod Knowlton
IBM Certified Advanced Technical Expert pSeries and AIX 5L
CompTIA Linux+
CompTIA Security+

 

[hirschaj]

You are right, deallocation of paging space is not accidental. I have seen people playing around with many features and not setting them back properly. Could it be early paging? Maybe, I have not had any experience in that arena either.

One thing I did want to mention about the numbers staying imbalanced. The Kernel will not clear out paging space until it needs to for some reason. I have seen an issue (at a few companies now) where ALL of the information in RAM gets paged out at once. It is a very rare occurance but I have heard of it happening at 3 or 4 companies on database servers (1 of them being my own). So on servers with large amount of RAM it forces you to make paging space at least equal to the size of RAM plus a little extra to keep on working.

That is very scary when you have a system with a ton of RAM. Unfortunately IBM is aware of this issue but I have not heard of a fix for it as of yet.

You make some good points Rod, it's nice to have these types of conversations with people who know what they are talking about


Jim Hirschauer

http://www.aixexpert.com

 

[Midrange]

Rod,

I have checked the /etc/environment file and i didn't see any PSALLOC parameters. DO you recommend to put that parameters on that file?

thanks.