Interpretation of wafl measurement 'hotspots'

[kevingraham]

Given a sample reallocation measurement:

    [wafl.reallocate.check.high:info]: Allocation check on '/vol/vol0' is 2, hotspot 31 (threshold 3), so will reallocate.

The overall volume measurement (2, in this case) is well documented and generally understood, but I've yet to find a definition for the 'hotspot' value (31 in above example). My understanding is that WAFL measurement calculations don't factor in usage patterns, so the tradition storage lexicography wouldn't seem to apply.

What is a 'hotspot' in this context? What factors can be done to alleviate them before getting to the reallocate?

In the sample, this is a healthy volume w/ reasonable utilization levels (there are no routine/scheduled reallocates on this vol, so that's a "natural 2") so WAFL seems to be doing a good overall job of managing layout w/ this exception.

Someone went to the effort to calculate this value and display it, so presumably there's value in the datum...

[Darkstar]

I have not found anything official on this but my guess is that it's some kind of "worst case" fragmentation. I.e. on the whole volume, the (averaged) fragmentation is 2, but the worst case fragmentation encountered (as seen over some kind of suitable window) is 31.

This is just my interpretation, I'd love to hear something official on this topic

-Michael

[kevingraham]

AdvUni-MD wrote:
my guess is that it's some kind of "worst case" fragmentation. I.e. on the whole volume, the (averaged) fragmentation is 2, but the worst case fragmentation encountered is 31.

Agreed, that's been my working assumption as well. Accepting that though, I'm still not clear on what factors contribute to the ugliness since on a whole we can see that WAFL is doing a "near perfect" (by its scoring) layout job. I would expect a sensible result to be marginal overall layout number combined with a nasty hotspot, such as 'measured is 8, hotspot 20'.

[ekashpureff]

I've done some testing for this.

I think it's a measure of something we refer to as 'vertical fragmentation'.

It's when you have data that's contiguous, but bound to a single or few spindles.

Vol0 on any filer will probably show up with this, given that it is usualy laid out on

only one disk at install time - before you grow aggr0.

I was just looking at a simulator. vol0 on the aggr0, which only has one data disk.

It shows an index of 1 but hot-spots of 24.

nfsflex is a volume on aggr1, made up of all the simulator disks.

It shows index of 1, with no hot spots.

Mmmk... more testing done.

In the example below vol0 was the original vol0 on the sim. vol0b was created after

growing out aggr0 with a bunch more disks, then ndmpcopy of vol0 data...

sim1> reallocate measure /vol/vol0
Reallocation scan will be started on '/vol/vol0'.
Monitor the system log for results.
sim1> Tue Oct 19 20:40:41 GMT [wafl.scan.start:info]: Starting WAFL layout measurement on volume vol0.

sim1> Tue Oct 19 20:41:09 GMT [wafl.reallocate.check.highAdvise:info]: Allocation check on '/vol/vol0' is 1, hotspot 22 (threshold 4), consider running reallocate.

sim1> reallocate measure /vol/vol0b
Reallocation scan will be started on '/vol/vol0b'.
Monitor the system log for results.
sim1> Tue Oct 19 20:41:45 GMT [wafl.scan.start:info]: Starting WAFL layout measurement on volume vol0b.
Tue Oct 19 20:42:04 GMT [wafl.reallocate.check.value:info]: Allocation measurement check on '/vol/vol0b' is 1

The hotspot index is an indication of spindle bound data.

At your service,

Eugene Kashpureff

Message was edited by: Eugene Kashpureff