expanding system dump area

[Mike Tufariello]

I want to expand my system dump area. It currently resides on /vg00/lvol2 at a size of 1024 MB. I want to increase it in size to 5 GB.

Is this as simple as using the lvextend command? Are there any special arguments I need to include because this is system dump?

Also because there is no file system mounted do I need to run the fsadm command? Or is there some variation of it just for system dump?

Can I do this while users are logged on? Or is this a system dedicated process that needs a re-boot?

Are there any other gotchas ?

Thanks,

[IT_2007]

you can't simply do lvextend since it needs strict and contiguous logical extents. You can create another logical volume as swap so that it can be used as secondary swap. Even you can specify to use it before using lvol2 in /etc/fstab like this

/dev/vg01/lvol1 .. swap defaults 0 0

last "0" specifies to use it first.

[IT_2007 · Accepted solution]

you can't simply do lvextend since it needs strict and contiguous logical extents. You can create another logical volume as swap so that it can be used as secondary swap. Even you can specify to use it before using lvol2 in /etc/fstab like this

/dev/vg01/lvol1 .. swap defaults 0 0

last "0" specifies to use it first.

If you still want to increase lvol2 from vg00 then you have to re-ignite server and when recovering it you can expand it.

[A. Clay Stephenson]

I am a very strong believer in the concept that swap and dump areas have absolutely nothing to do with each other and should thus be separate areas. If this were me, I would define additional LVOL's and run multiple lvlnboot -d commands to define more than one dump space. Dump should never be mirrored and swap should always be mirrored. Having separate dump and swap areas also allows your machine to recover more quickly because now you can disable savecrash as there is no need to capture and compress the dump image to a file because it will not be overwritten by swap usage.

[Mike Tufariello]

Thanks to both for the quick response. After reading and understanding each response the following thoughts arise;

I agree that swap and dump should be separate. And now understand that dump has to be contiguous. If I can not make the lvol2 on vg00 contiguous without a major effort, can I create new lvols on vg00 and designate them as swap? My vg00 is mirrored, so I am concerned and do not understand the comment that dump should not be mirrored. And why should I define multiple dump spaces rather than one at the size I need?

How can I tell how much contiguous disk space I have available on vg00 ?

Thanks,

[IT_2007]

Dump only used to save crash dump when system crashes and swap is to hold processes while system running and it is like a scratch pad. So you need mirror for Swap and no need to have mirror for Dump.

To check contiguous logical extents, vgdisplay -v vg00 and see how many extents you have per disk and PE size.

[Mike Tufariello]

So I've got 3378 free PEs @ 16 MB per PE for a total of 27 GB when mirror is taken into account. Does this mean I can create my 5 GB dump space from the 27 GB free and know that it is contiguous? By default is will end up being mirrored. Do I really care or can I prevent the new dump space from being mirrored?

BTW, I came up with 5 GB dump space based on the /sbin/crashconf -v formula of total pages in dump times 4 KB times 1.25.

Thanks,

[IT_2007]

So I've got 3378 free PEs @ 16 MB per PE for a total of 27 GB when mirror is taken into account. Does this mean I can create my 5 GB dump space from the 27 GB free and know that it is contiguous? By default is will end up being mirrored. Do I really care or can I prevent the new dump space from being mirrored?

BTW, I came up with 5 GB dump space based on the /sbin/crashconf -v formula of total pages in dump times 4 KB times 1.25.

=======================================
Yes. You can create 5GB of dump space. Even if you mirror it won't be any problem. If all other logical volumes are mirrored then go ahead mirror dump too. When you create logical volume using lvcreate use "-C" option to make it contiguous.

[A. Clay Stephenson]

No. It's dumb to mirror dump because it is a crapshoot as to which of the physical devices the dump will actually be written to. With only one choice (ie no mirror) then it's a constant. The smart play is to create 2 smaller dump areas of equal size so that the primary and mirror disk end up being identical and no space is wasted.

[IT_2007]

I agree with Clay. That make sense to make it equal on both sides.

[Mike Tufariello]

OK. So if I create a dump space of 1/2 the 5 GB, 2.5 GB, I will have the primary of 2.5 GB and the mirror of 2.5 GB. Are you saying that if I have a crash the system will know to use the primary and then use the mirror? I thought the dump space had to be contiguous? Is the mirror contiguous with the primary by default? I am not sure I understand your point about the smaller areas of equal size.

Thanks,

[IT_2007]

He mean to say that create two dump areas instead of one.

Like Dump1 size of 2.5GB on Primary disk and 2.5GB of Dump2 on alternate disk so that they would be used to collect dump.

If you look at /dev/vg00/lvol2 also acts as swap and dump .. After configuring like above, you would have 2 more dumps.

[Mike Tufariello]

Like Dump1 size of 2.5GB on Primary disk and 2.5GB of Dump2 on alternate disk so that they would be used to collect dump.

If you look at /dev/vg00/lvol2 also acts as swap and dump .. After configuring like above, you would have 2 more dumps.

Sorry. So now I'm more confused. Dump1 on the Primary disk will mirror by default. What do you mean by alternate disk for Dump2? I thought the disk space had to be contiguous?

[IT_2007]

Sorry for the confusion. I mean to say new dumps as Dump1 and Dump2.

you create them as dumps only and they are two logical volumes and each one is contiguous.

System dumps will be saved wherever dumps defined and don't have to one big size logical volumes.

Hope this makes clear to you.

[A. Clay Stephenson]

Do a man lvlnboot and look under the -d section. You will find that you can run the command multiple times to define multiple dump areas and as the 1st fills up, the 2nd is used, and so on.

On systems with large amounts of physical memory (e.g. 32GiB or so), something like 25% of the total is sufficient for a dump area because almost certainly all the significant data will be captured in 8GiB.

[Mike Tufariello]

Right. I determined that based on my configured memory and crashconf -v that I need 5 GB. But I thought we were discussing the issue of letting dump volumes mirror? And ways around that.

I not sure I understand the difference of creating 1 lvol of 5 GB or 2 lvols of 2.5 GB each.

I did review the lvlnboot command and see how to make he lvols contiguous, etc.

Thanks,

[A. Clay Stephenson]

No, dump volumes should never be mirrored (unless they also serve as swap volumes and then they should always be mirrored).

If you want to configure separate dump and swap areas (and I 'll take the simplistic view that you are mirroring on identical physical disks) then all the LVOL's in /dev/vg00 should be mirrored except what ever LVOL's you create to be the dump area's. For that, if for example, you wanted 5GiB total of dump space, you would create 2 2.5GiB LVOL's (with different names and allocate each one from separate PV's). You then run lnlvboot -d twice and specify these dump LVOL's.

[Mike Tufariello]

I understand that we want to keep dump and swap separate. I understand that you do not want to mirror any dump space. And you are correct with the simple view. I am mirroring vg00 on 2 identical disk drives, from PV /dev/dsk/c0t6d0 to PV /dev/dsk/c3t6d0. If I create an dump lvol of 2.5 GB on vg00 on PV /dev/dsk/c0t6d0 will it not be mirrored by default to PV /dev/dsk/c3t6d0 ?

How do I stop the mirroring from PV /dev/dsk/c0t6d0 to PV /dev/dsk/c3t6d0 ?

I must be missing something.

Thanks,

[IT_2007]

I got your point. It is not possible to create a new logical volume on to mirror disk since it is already mirrored. I don't know if you can do this way or not. I never tried but logically should work.

1. Create one logical vol without mirror (lvcreate) on primary disk and make it as dump (lvlnboot -d)
2. Reboot system with alternate disk and do the same thing above.

[A. Clay Stephenson]

Your fallacy is thinking that the mirroring is done at the PV (disk) level when it is really done at the logical volume. You do an lvextend -m 1 on every LVOL except for the dump LVOL's. It is important that there are two separate and distinct LVOL's for dump (e.g. /dev/vg00/lvol9 and /dev/vg00/lvol10).

[Mike Tufariello]

So I think I get it now. In my case, however, I want to create 2 new lvols, so I would use lvcreate, not lvextend, without the -m option, so that the new lvols are not mirrored.

However, 2 more questions arise.

1) If I have a system crash that requires booting from the alternative disk, which in my case is the mirror disk, I would not have any dump space to write to.

2) I already have lvol2 set up as a dump/swap on vg00. Can I or should I safely remove lvol2. Or should I just change it in SAM to be the last device the dump should look for.

Thanks,

[A. Clay Stephenson]

Who says if you boot from the alternate that you will have no dump area? You run the lvlnboot -d command multiple times and as many of these LVOL's are actually available will be used.

No, you cannot remove a swap volume while the machine is running but you can define dump volumes "on the fly".

I should note that this is all hypothetical in my case because I literally cannot remember the last time I had a production crash.

[Mike Tufariello]

Thanks for all your help and patience....