"Glenn Dekhayser" <gdekhayser(a)voyantinc.com> writes:
> Netapp recommends that you keep the fileysystems at 80% or less. Otherwise,
> WAFL starts to work very hard finding places to put data, and you'll
> see CPU spikes, High disk utilization, and degraded performance.
while Jeff Burton <jburto1(a)yahoo.com> writes:
> We have a FAS940C that has 2 trad volumes (5TB & 3TB, 144GB/FC drives) that are
> always around 95%+ full because of funding and timing issues. Only when we run
> out of space do we start deleting snapshots, otherwise there are always a few
> hundred GB of snapshots. We've been running over 90% for the past year and
> have never had an outage or major hiccup...just a few disk failures. We
> typically have 4k-5k active CIFS connections and around 100 NFS connections.
> Clients have never noticed a performance problem, and our CPU utilization
> hovers between 30% and 50%. So maybe we do not notice a problem because there
> is more than enough CPU cycles left over for DOT/WAFL to deal with the full
> file system.
which seems a clear case of YMMV !
It's worth remembering that ONTAP already imposes a 10% reserve on the
filing system size (including snapshot reserve); or on the sum of sizes,
for flexible volumes in an aggregate. A "100% full" tradvol/aggregate is
actually only using 90% of the blocks on the discs. Or looked at another
way, the difference in congestion between "90% full" and "100% full" is 2:1,
not infinity:1.
"Blake Golliher" <thelastman(a)gmail.com> writes:
> A filer can be written to over 100% space utilization, it'll just keep
> growing. I've seen filers go up tot 107% space utilization before.
and "George, Andrew" <georgea(a)anz.com> replies:
> Interesting
> We run a fair few filers in the 97%-99% arena, mainly as CIFS NAS devices.
> Every time I've seen it hit 100% 0 bytes available CIFS has refused to
> save anything.
That's my experience, with NFS. I would be interested (from a theoretical
point of view!) to know how Blake gets the space utilisation that high.
(Snapshot usage vs reserve as shown by "df" can go over 100%, of course.)
--
Chris Thompson
Email: cet1(a)cam.ac.uk