Re: nutanix

From: Jeremy Schneider <jeremy.schneider_at_ardentperf.com>
Date: Wed, 10 Sep 2014 09:05:33 -0500
Message-ID: <CA+fnDAYncTS9rTtvmtx0kdWqOqwVt4D_rrrAYtntO+e4rj=B-w_at_mail.gmail.com>



On Wed, Sep 10, 2014 at 7:03 AM, max scalf <oracle.blog3_at_gmail.com> wrote: >
> You definitely bring up a great point with regards to CBO, but wouldn't
that apply to exadata as well?

Yes and no. I don't think that the costing algorithms will adjust the cost of a table scan or index scan based on caching, and the CBO doesn't have any awareness of storage indexes or flash. However on exadata, the CBO will consider access paths that aren't available on non-exa platforms (smart scan). Also, the runtime engine will consider cache information - for example, the direct path decision takes into account what percentage is cached in the SGA and this may include database smart flash cache (in the compute node), though I'm not sure about that. and FYI, if you're on solaris or OEL then you can use server-side flash cache with any enterprise edition install (non-exadata).

Essentially, it's different to the extent that the database software is coded to leverage storage tiering. Imagine configuring a tiny SGA and a huge OS filesystem cache, and hoping that the OS filesystem cache will keep all your objects in memory. It will certainly cache stuff - but the SGA does a much *better* job of keeping the right data in DRAM because it has a lot more information to accurately predict which data will be accessed next.

> wouldnt that same case apply to traditional model as well when we have
SAN in place, as the data is caches in SGA first and then most probably in the SAN Cache and different tires of storage on SAN?

same idea with SAN cache - there are only a few cases where they give significant benefit because in general they are poor at predicting which blocks will be accessed next (compared to the SGA). james morle's old (and legendary) sane san paper discussed this and contrary to the storage rep sales pitch, it's still true today.

> according the Nutanix rep's they have big enterprises running oracle/SAP
on there system and have not seen any major issue.

then they should be able to provide reference customers who can talk with you directly. talk to a CIO and some engineers - without the nunatix rep around so that they can talk freely. dig deep into how they are using nunatix, how hard they've pushed the database on the platform, where it excelled and where the biggest challenges were.

> http://stevenpoitras.com/the-nutanix-bible/

really cool link - thanks for sharing!

-Jeremy

--
http://about.me/jeremy_schneider

--
http://www.freelists.org/webpage/oracle-l
Received on Wed Sep 10 2014 - 16:05:33 CEST

Original text of this message