We use Pure Storage. We have a *very *busy system that their FA450 had some
trouble keeping up with. We would sometimes see a parallel index creation
or expdp from dev cause a latency spike in the array that would result in a
spike in log file sync waits on prod. Pure says this was due to throwing
more I/O than their NVRAM can triage, so things back up.
We've just upgraded to an //m70 this past week. Not a lot of time to judge
but it seem to be doing great so far. We definitely were pushing the limits
of the smaller system. Hopefully the //m70 better handles what we throw at
it. Pure Support did say that we were in the top 1% of their clients in
terms of I/O activity. If the //m70 keeps performing well we are
considering adding another to each datacenter. Snapshot replication between
arrays is also great if you have the network bandwidth for the given DB
size.
The thin snapshot cloning is the greatest thing ever. Cloning our prod DB
for dev or staging went from 2-3 days to 15 minutes. We built a script
around it so spinning up new clones can be in a few minutes once the
volumes are provisioned and assigned to the DB hosts. The browser UI is
also very nice, and REST API is great for scripting.
The one gripe I have with their support team is that they were quick and
insistent on blaming Oracle or our DBA team for the performance problems in
the FA-450 array, even after we kept clearly pointing out that the issues
happening across DBs show that it can't be. And the fact that I once caused
a latency spike by running an "rm" on a set of datapump files on a volume.
Just today we had a call about volumes unmounting during "online"
maintenance causing DBs to crash and they kept asking what version of the
database we were using. That's like asking what model car you drive if you
hit a pothole in the road and get a flat tire.
We also used to have a ZFSSA (7420). Those here that are familiar with my
tweets and emails know that I would not recommend it for running databases
on. If you use it to hold backups, fine. But we had nothing but horrible
performance trying to run even our staging/test databases on. First it was
problems with infiniband and DNFS that were due to bugs in the OFA kernel
module that are fixed in OEL but Oracle refused to share with RHEL (we use
RHEL). Then we'd see problems where the NFS mounts would suddenly hang and
the databases would crash. Every part of the network stack was poked and
prodded and nothing was ever found.
The most frustrating part of it all was Oracle Support. Their selling point
was that having Oracle for database and storage meant less juggling for us.
This was not the case as we'd have to re-explain the situation every time
an SR got handed off to a new tech. Eventually we just had to tell our
account manager to cut the bullshit and so everything flowed through him.
In the end we had to negotiate a return and I'm pretty sure Oracle just
sent them to scrap.
I'm sure there are folks out there that have had great experiences with the
ZFSSA and live databases (their sales people and PMs insist so) but it
certainly was a huge rotten lemon for us. The real thumb in the eye was
when they tried to steer us to an Exadata to replace it.
Don.
On Wed, Apr 20, 2016 at 2:27 PM, Jeff Chirco <backseatdba@xxxxxxxxx> wrote:
Anyone use or used Pure Storage for your database storage? We currently
have NetApp and it is coming on a hardware refresh so we are looking
around. We've looked at the ZFS which seems really nice, a little more
expensive than we though but great for Oracle database. Now my system
admins are looking at Pure Storage as a possibility as well. From what I
read it seems like a all flash system and apparently hardware freshes are
built into the support contract. Let me know if you have any onions.
Currently we are running Oracle on Windows Server 2008r2 with 11.2.0.4 but
moving all over to Oracle Linux and eventually 12c.
Thanks,
Jeff
