boot, swap , root each on separate disks - is it possible ? [amount of device swap]

[Mariappan_B]

Hi Colleagues,

                          During OS install / reimage ,  Is it possible to have HPUX boot volume on one disk , root volume on another disk and swap on another disk ( all these 3 disks will be part of vg00 , obviously other filesystems will be there )

scenerio 1:

vg00 with 3 physical disks :

boot volume on one disk

root volume on another disk

swap on third disk

scenerio 2:

vg00 with 2 physical disks:

boot and root volume on one disk

swap on another disk

Is it possible to have like mentioned above ? I already know that boot,swap,root each should have contiguous space on disk . In the above scenerios these 3 volumes will still have contiguos space on it's own disks .

Thanks for your view on this

Thanks & Regards,

Mariappan

[Patrick Wallek]

I have never heard of anyone trying anything like this before.  It might work, but I'm not sure it would be a supported configuration.

Why would you want to do something like this?  I don't know what the benefits would be here.  I really do not see the point. 

[Torsten.]

What should be the use of this?

Usually the first LVOLs 1...3 are at the beginning of the boot disk.

What do you want to do with the remaining space on these disks? Waste the space?

If you are concerned about the swap, additional swap can be configured on other disks too.

[Matti_Kurkela]

> Usually the first LVOLs 1...3 are at the beginning of the boot disk.

On PA-RISC, that was a requirement of the boot loader; with Itanium, this restriction could probably be lifted without too much development effort.

Perhaps the original poster has a system with a huge amount of RAM, and the old swap sizing rules would make the swap too big to fit on the same disk with the rest of the system?

To that, I would say two things.

First: the old "swap = 2x RAM" rule is obsolete. If you have hundreds of gigabytes of RAM, and the swap area is sized according to that obsolete rule, the system performance will drop to unacceptable levels long before the swap is 100% used.

If you have the pseudo-swap enabled (kernel parameter swapmem_on), all the system RAM can be fully utilized even if your total amount of swap is less than (<amount of RAM> / 4).

With all modern versions of HP-UX, the swapmem_on tunable has been enabled by default; with 11.31, the tunable no longer exists and the functionality cannot be disabled.

Second: even if you have to have lots of swap, there is no requirement to have all of it as primary swap (/dev/vg00/lvol2).

You could make the primary swap small enough to fit onto the normal system disks, and have the necessary amount of secondary swap allocated elsewhere. At worst, you might have to configure a separate crash dump disk. I've seen a recommendation for sizing the crash dump area: it was 2.5 GB + (<the amount of system RAM>/10).
 Anyway, the "crashconf -v" command should give you a better estimate of the required size of the crash dump area.

[Joe Ledesma]

(Side topic about how much device swap should be configured)

@Matti_Kurkela:

Am not disputing that:

> First: the old "swap = 2x RAM" rule is obsolete.

But sharing an example of a major application, SAP, for which the recommendation is still made.

In Swap/Paging Space in SAP on HP-UX Best Practices - Part 2 it has:

> In general device swap configurations of 1.5x or 2x RAM have proven appropriate for the majority of SAP installations. The recommendation is to set device swap to 2x RAM (minimum 20GB). 

And refers to a technical SAP note (requires logon) but reposted here that concludes:

> The general recommendation is to configure device swap to 2x RAM. In some cases it may be possible to tune this lower, but a minimum of 0.5x RAM should be maintained. As a rule of thumb the total line of "swapinfo -tam" should not exceed approx. 80% when the HP-UX OS is running the configured SAP systems under planned workload.

Am planning the config for device swap for a system for SAP with:

> If you have hundreds of gigabytes of RAM,

[Patrick Wallek]

In my opinion that is because the people at SAP are clueless.  In a system with 128 GB of RAM, having 256 GB of swap space is absolutely ridiculous.

Pseudo swap (swapmemon=1) should be fine so that you can actually use all of your RAM.  Having to have actual device swap of 2 x RAM, when no modern system should EVER page out significantly, is an ineffective use of disk space.

[Torsten.]

This old rule was probably valid when systems came with a couple of megabytes of RAM years ago. Now you can have up to 4 TB of RAM, would a swap of 10TB make sense? No.