Originally posted by: MimiX

a little paging question.

Environment:

AIX 5300-2 64 bit on P570 (LPAR), DPSA paging algorithm
2 IBM internal disk (16 bit LVD SCSI) 73.4 GB (mirrored rootvg)
2 vpath SAN attached to a 2105-800 20GB each one (user data)

From performance point of view, is better a mirrored paging space in rootvg or one copy in rootvg, or a one copy on ESS disk?

Finally ESS disk are faster then internal disk?

TIA
MimiX

Originally posted by: nagger

This is an "old chestnut" type question best discussed with a cold beer.
There is no one correct answer.

Most people protect their paging spaces in these days of cheap disks.
Not mirroring or RAIDing via the ESS (or whatever disk subsystem) means that a paging space disk error will halt your system.

There was a time when specific disks were used for paging.
Now people tend to avoid bad paging by using larger memory - which gives you a larger buffer and more time to free memory up memory in advance. Also large system actually page most of the time - it is just they way they work and users/wokload change process working pages.

Most people now go for the "hose it all about" approach i.e. placing some paging space on lots of disks with the aim to make sure that if you hit a period of bad paging there are lots and lots of disks helping out to get through the peak as quickly as possible.

Which is faster SCSI or ESS? - it depends. There are thousands of ways of doing I/O and caching the data for efficiency.

Originally posted by: MimiX

Cold beer open the mind.

Thanks nagger, this confort my choice to have paging in ESS (RAID array).

Originally posted by: VirtualGreg

I have customers that are booting from their ESS and some that boot from virtual disk that is looked after by a VIOS LPAR and in turn the boot disk is either an LV on an ESS LUN, or the ESS LUN itself mapped in whole to the client LPAR. No issues. Your rootvg I/O rates should be quite small. I've often thought this was a wierd question - if the ESS disk is good enough for high performance database spindles, rootvg is a walk in the park!

Another paging question I hear a lot: How much do I need? The only reasonable answer these days is: Don't run out! There are all sorts of ways to monitor paging space utilization including the no charge option within AIX: RMC.

Originally posted by: MimiX

plz, if u can, explain the phrase: "There are all sorts of ways to monitor paging space utilization including the no charge option within AIX: RMC."

TNX.

Originally posted by: gcorneau

Here's a link into the AIX documentation that discusses RMC:

http://publib.boulder.ibm.com/infocenter/pseries/v5r3/topic/com.ibm.help.rsct.doc/rsct_books/rsct_admin_guide/bl5adm1036.html

Originally posted by: SystemAdmin

Using RMC:

What if you're not running a cluster? I've got H80's and 6H1's.

Can RMC be run on its own? Or is RSCT required?

Is it worth the effort? Install, configuration, overhead.

Does it provide more than topas, or nmon commands?

From the perspective of monitoring paging space usage more effectively than
repeted "lsps -a"

Originally posted by: nagger

There is nothing to install.

Yes, you have to configure it so that it reports in the way you want.

There is near zero overhead - when the kernel notes you going over the threshold it fires the event, so there is no wasteful polling or command to run.

I hope, this will encourage you and others to give a this advanced and free AIX function a serious go.

Originally posted by: niella

Seems like "it depends" is the safe answer.

My take on this is that it is simpler to create paging on internal (or direct-attached) disk and use a RAID-1 mirror (under the assumption that controller failures happen seldom).

There are more considerations in terms of points-of-failure if you were for example using a switch attached storage device. In a simple setup a single cable failure (or a switch reboot) would result in an outage.

Other factors would be the RAID level of the storage where you want to create your paging on, and how redundant your paths are. RAID-5 would probably be best avoided for paging and is sometimes not as safe as is generally presumed - if you do some investigation on BST (Bad Stripe Table) entries, you'll realise that the reliability RAID-5 arrays depend on a few important factors. (Check IBM's serveRAID troubleshooting area) Also, RAID-5 performance penalties are more severe for disk writes than RAID-1.

Regards,
Niel

Originally posted by: MimiX

Yes, is right! the entire infrastructure environment and single point-of-failure must be considered, but if all of them are ok, the question is .....how faster i can accomodate (tuning on the configuration) pgspin pgspout from the point of view of the user response time? i.e. when a user application peak is reached and system spend 60% in iowait for paging what i can do?

Well the first think i have done is to move paging on ESS (Raid-5 array), and the response time result more confortable.

De facto, paging on external disks (ESS Raid-5 array) are faster than internal!