Mega- swap space question

[Sanjay_6]

Hi,

you cannot put filesystem swap on cluster VG. So you have to find an internal or external disk not used by the cluster VG. The cluster VG gets activated by the cluster packge and hence your swap will not get activated if you put the same on cluster VG/LV.

Also take a look at these threads for more info,

http://us-support.external.hp.com/cki/bin/doc.pl/sid=7e88f7890d34e7560e/screen=ckiDisplayDocument?docId=200000024603528

http://us-support.external.hp.com/cki/bin/doc.pl/sid=7e88f7890d34e7560e/screen=ckiDisplayDocument?docId=200000054497857

Hope this helps.

Regds

[John Bolene]

You need more hardware, 256MB memory is not near enough to support that many users. 512 would be more like what you need.

You did not say what machine you have (or I did not find the info), I would suggest at least a D390 with dual processors and 1G memory for that many users.

[Santosh Nair_1]

1. This is normal. When a process starts up, it puts an image into physical memory and it also sets aside space in swap area called reserve. This is so that the process has somewhere to go if it ever gets swapped out. It would be safe to add another swap device.

2. The standard rule of thumb for swap is that it should be 1.5 to 2 time physical memory. Also, make sure psuedoswap is turned on (swapmem_on should be set to 1). Finally, you might want to decrease your buffer cache. By default its set to 50% which in almost all cases is way too high. Set it to 10-15% (dbc_max_pct). This will give you back some memory.

3. Just create a new lvol and assign it as swap (if you don't know how to do it, you can use SAM to do this). Also, make sure maxswapchunks is set high enough to accomodate the additional swap (maxswapchunk * 2M <= total swap).

4. Make sure that you don't create the swap device on a MC/SG shared volume. This would lead to huge problems.

-Santosh

[Krishna Prasad]

Do you have glance?
How much memory to you have on your box?

Also, we had the exact error message the other day....It was not an OS/MEM/SWAP problem. We fixed it by increasing the Oracle SGA.

What most likely happened was the SGA was full and needed more Oracle need more memory and could not get it.

The problem you may run into after increasing the SGA is that less physical memory will be left after Oracle starts. You need to watch your free memory as users start to log on to the box.

You may need to add memory to the box. However, increase your SGA size regardless.

[Wodisch]

Hello Manuel,

1) you need more RAM
2) you need more RAM
3) you need more RAM
4) you should consider using "pseudo swap"
5) you need more RAM
6) you should consider modifying your maximum buffer cache size to something small
7) you need more RAM
8) you should consider using the "Online JFS" for the disks with the Oracle-Data-Files and then use the mount options "datainlog,log,mincache=direct,convosync=direct" for those file systems (well, only if you store there the data files ONLY - nothing else!!!)
9) you need more RAM
10) consider stopping everything you do not use, e.g. NFS

Just my $0.02,
Wodisch
PS: you need more RAM :-

[Jim Turner]

Hi,

I'm afraid you're in the classic position of trying to put ten pounds of sh...stuff in a five-pound bucket. Any changes to your current config will simply move the problem from one place to another.

You're only real solution here is to get more RAM and more disk.

Cheers,
Jim

[A. Clay Stephenson]

Hi:

Plan B. Is there a chance that your users are loading tons of common data and that your application could let them access a common shared memory segment? This would dramatically reduce your memory requirements and you application would scale much better.

[Manuel G]

Thanks for your responses.
Related with things you suggested:

*Sanjay:
I suspected that swap space can not be configurated on cluster VG so we had rejected this posibility. I can not open threads you linked, can I go there by other way?

*Jhon:
I agree with your machine suggestion but I guess our customer doesn??t. Our system is composed by 2 HP-9000 D230 servers with one 180 MHz. PA-RISC processor each. We are planning to increase memory to 512MB as you suggest.

*Santosh:
Great description of swap usage.
I??m not sure about pseudoswap, I??ll check it tomorrow.
Our dbc_max_pct and dbc_min_pct are set to 10 and 2, I think it??s OK, is it?
What??s the function of maxswapchunks? I don??t know its value exactly, I??ll check it tomorrow too.

*Ron:
Our sga is 200 MB more or less. I think increse it could not be a good idea with only 256 MB RAM. In order to checking memory usage we only know "dmesg" command, but it returns sysmap overflow errors ...

*Woodish
I absolutely agree with you about RAM, our customer begins to consider it;-).
How can I configure pseudo-swap?
Related to Oracle-Data-Files, they lay in external clustered disks, I prefer to try with your 1,2,3,5,7,9 suggestions.
We??ll review the system for searching useless processes.

*Jim:
We??ll try with memory increase but not yet with disk increase (our customer will not like this very much). In order to increase memory we should reduce some or several lvol/s for increasing swap as well, does lvreduce has any data loss danger? I guess I should backup chosen lvol first and then "lvreduce" it.

*Clay:
Great Idea Clay (you always give a different point of view), but we have already make several dinamic libraries for sharing data. Nothing else to do on this way.

- I think everybody is agree with the idea of increasing memory. The only thing that makes me be afraid is reducing some lvol and lose data and make the system unbootable or something worst. Any safe suggestion?

Thanks for paying attention.

[Sanjay_6]

Hi,

If you are using europe-itrc then replace "us-support" in the link with "europe-support". Your ID should be linked to a valid hp agreement that allows you access to itrc tech docs.

Hope this helps.

Regds

[Jim Turner · Accepted solution]

A make_tape_recovery (twice wouldn't hurt) and a couple of backups of the affected data are certainly in order before you reduce any logical volumes.

If your filesystems have been extended to fill the logical volumes (which they almost surely have been), then reducing the lvol is destructive. If you want to reduce operating system lvols, boot interactively from your make_tape_recovery, resize them, and then let the recovery run its course. If it's a non-OS lvol that you're reducing, you are best served by backing it up, blowing it away, and recreating it at the (smaller) desired size. You can then restore your data.

Cheers,
Jim

[John Bolene]

You may not need to increase disk swap, in the tunable kernel parms, do a swapmem_on 1, this will turn on pseudo swap.
If it is already on, you will need to increase swap to 1G after you increase memory to 512.

[Manuel G]

*Jim&Jhon

I will lvreduce lvol4-/home because it won??t grow much more. I think there is no OS problem on this lvol.

If pseudo-swap isn??t enable, is better enable it or increase swap?

Thanks.

[John Bolene]

Enable pseudo swap first, then see if you are still getting the out of swap space error.

You will then need to increase swap space.

Sounds like you are getting a handle on the problem.

[Jim Turner]

/home is probably the least impact of the vg00 lvols to reduce. Booting from a make_tape_recovery is not necessary, although you should go ahead and make the tape as a safeguard. When you're ready, log in as root and make sure everyone else is logged off. If not, terminate them off yourself (God, I love my job). Take a couple of good backups of /home using your weapon of choice (tar, cpio, fbackup, omniback, etc.).

Now unmount /home and lvremove the lvol. Create a new lvol at the desired (smaller) size, specify the same mount point (/home), and restore your data. Obviously, you'll need to make sure the new size still accomodates the data that is in /home right now.

Cheers,
Jim