Re: Is filesystemio_options relevant when the database is on ASM ?
From: Kevin Jernigan <kevin.jernigan_at_oracle.com>
Date: Fri, 17 Oct 2014 10:06:20 -0700
Message-ID: <54414C8C.4020301_at_oracle.com>
Direct NFS (dNFS) can lead to challenges for the underlying NAS infrastructure, because it eliminates the performance bottlenecks inherent in most OS implementations of the NFS client (what we usually call "kernel" NFS or kNFS). For example, dNFS will allow each Oracle process to make a direct connection to the NAS server, rather than funneling all I/O requests through a single connection. This means that the NAS server(s) need to be able to handle the increase in I/O traffic that the database will request from the storage.
Date: Fri, 17 Oct 2014 10:06:20 -0700
Message-ID: <54414C8C.4020301_at_oracle.com>
Direct NFS (dNFS) can lead to challenges for the underlying NAS infrastructure, because it eliminates the performance bottlenecks inherent in most OS implementations of the NFS client (what we usually call "kernel" NFS or kNFS). For example, dNFS will allow each Oracle process to make a direct connection to the NAS server, rather than funneling all I/O requests through a single connection. This means that the NAS server(s) need to be able to handle the increase in I/O traffic that the database will request from the storage.
So, it's not a flaw in dNFS that creates problems for some NAS setups - there are often performance limitations in the NAS server that were masked by kNFS limitations, and dNFS removes those limitations. dNFS does extremely well in environments that have the right setup to handle higher IOPS etc - if the application workload on top of the database requires a certain level of I/O throughput and latency, it's much easier to get to those levels with dNFS than with kNFS.
-KJ
-- Kevin Jernigan Senior Director Product Management Advanced Compression, Hybrid Columnar Compression (HCC), Database File System (DBFS), SecureFiles, Database Smart Flash Cache, Total Recall, Database Resource Manager (DBRM), Direct NFS Client (dNFS), Continuous Query Notification (CQN), Index Organized Tables (IOT), Information Lifecycle Management (ILM) +1-650-607-0392 (o) +1-415-710-8828 (m) On 10/17/14, 8:13 AM, Hameed, Amir wrote:Received on Fri Oct 17 2014 - 19:06:20 CEST
>
> We started using DNFS about 1 ½ years ago and have been bleeding ever
> since. One of the issues is with the NAS technology from our vendor
> which we have found unstable. The IO timings fluctuate too much and
> the 10046 traces can easily prove it. The other issue is most likely
> with our network infrastructure, which we were told initially both by
> our data center folks and by our network vendor that we did not need a
> separate dedicate network for the private NAS traffic and that the
> existing switches would provide the QoS that is needed but now both
> are proposing a dedicated set of switches to try and resolve the issue.
>
> DNFS works fine for environments that are not very demanding on the
> IOs and latency but based on our experience, it does not do well with
> high IOPS and low latency requirements.
>
> *From:*Don Seiler [mailto:don_at_seiler.us]
> *Sent:* Thursday, October 16, 2014 4:20 PM
> *To:* Kevin Jernigan
> *Cc:* Hameed, Amir; fuzzy.graybeard_at_gmail.com; oracle-l_at_freelists.org
> *Subject:* Re: Is filesystemio_options relevant when the database is
> on ASM ?
>
> For what it's worth, we have the 7420 (two of them). My criticism of
> them doesn't come lightly. It's been a long year+ and the fact that
> these problems have kept us from migrating to them has resulted us
> having to deal with other emergencies from our current aging storage
> that we hoped to be off of nearly a year ago.
>
> Don.
>
> On Thu, Oct 16, 2014 at 3:12 PM, Kevin Jernigan
> <kevin.jernigan_at_oracle.com <mailto:kevin.jernigan_at_oracle.com>> wrote:
>
> Don,
>
> Based on your experiences I think anyone would understand your
> perspective. My areas of responsibility within Oracle are all
> within the database team, so I am not an expert on ZFSSA (ZS3)
> storage, but I will contact the ZS3 product team to see if they
> have any comments.
>
> Thanks for your feedback,
>
> -Kevin J
>
> --
>
> Kevin Jernigan
>
> Senior Director Product Management
>
> Advanced Compression, Hybrid Columnar
>
> Compression (HCC), Database File System
>
> (DBFS), SecureFiles, Database Smart Flash
>
> Cache, Total Recall, Database Resource
>
> Manager (DBRM), Direct NFS Client (dNFS),
>
> Continuous Query Notification (CQN),
>
> Index Organized Tables (IOT), Information
>
> Lifecycle Management (ILM)
>
> +1-650-607-0392 <tel:%2B1-650-607-0392> (o)
>
> +1-415-710-8828 <tel:%2B1-415-710-8828> (m)
>
> On 10/16/14, 12:59 PM, Don Seiler wrote:
>
> Of course, it's entirely possible that the two ZFSSA units
> that we received are the only two lemons off of the assembly
> line. That would explain by the ZFSSA support techs had so
> much trouble finding/fixing the problems (most of which are
> still not fixed).
>
> Last summer I couldn't have been more excited to get these
> units installed and start using them. But it was clear from
> the start that things were going wrong, and it was a series of
> problems from the start in various components of the ZFSSA.
> Even better was the fact that the two units each had unique
> problems, failing differently than their counterpart.
>
> So, given the problems we've seen, considering both the
> quantity and severity, and how completely unimpressed we were
> with the "one-stop shop" for support that was one of the big
> selling points, we can in no way consider moving our
> production databases onto it, and are looking for alternative
> storage to remove our staging and unit test databases off of
> it. Even if/when the open bugs get fixed and someone discovers
> the cause of the NFS hangs, the trust is completely gone in
> these systems.
>
> Don.
>
> On Thu, Oct 16, 2014 at 2:41 PM, Don Seiler <don_at_seiler.us
> <mailto:don_at_seiler.us>> wrote:
>
> We've had nothing but problems with them from installation
> misconfigurations to hardware failures (two silent NIC
> failures in addition to the disk failures we get warnings
> about, and one instance of the entire appliance IO module
> crashing). The last few months we'd see intermittent NFS
> hanging for 5-6 minutes to all mounts from the ZFSSA,
> resulting in those databases crashing. That is the most
> worrisome. If the storage can't stay online, then we have
> bigger problems than worrying about speed.
>
> Add on to that the less than stellar support
> recommendations we've gotten flip-flopping around
> Infiniband recommendations, write-bias setttings, bonding
> configurations, etc. and it's been a complete nightmare
> that's left us still on our old storage that is starting
> to fail. I have zero confidence in the ZFSSA (at least the
> two machines that we've been sold) to run our database.
> This is 14 months after installation and they're still not
> in production and never will be.
>
> You say you have many customers on it, but we found this
> hard to believe given all of the bugs that we tripped over
> along the way, including a couple that were apparently
> discovered by us. We would have expected support to give
> us a heads-up about all of the needed patches if they have
> so many customers that have done the same thing. Other
> than the recommended Direct NFS patches MOS note, we've
> been basically stumbling around in the dark. At various
> points, support suggests patches that only *might* fix the
> problem ... as if they aren't sure themselves. And those
> patches don't ever fix the problem.
>
> Don.
>
> On Thu, Oct 16, 2014 at 2:18 PM, Kevin Jernigan
> <kevin.jernigan_at_oracle.com
> <mailto:kevin.jernigan_at_oracle.com>> wrote:
>
> Don,
>
> Why do you recommend not using ZFSSA for live
> databases yet? We (Oracle) have many customers using
> ZFSSA for live production environments - including our
> own IT department, with >200PBs of ZFSSA storage in
> place for applications that support all aspects of our
> business...
>
> -Kevin J
>
> --
>
> Kevin Jernigan
>
> Senior Director Product Management
>
> Advanced Compression, Hybrid Columnar
>
> Compression (HCC), Database File System
>
> (DBFS), SecureFiles, Database Smart Flash
>
> Cache, Total Recall, Database Resource
>
> Manager (DBRM), Direct NFS Client (dNFS),
>
> Continuous Query Notification (CQN),
>
> Index Organized Tables (IOT), Information
>
> Lifecycle Management (ILM)
>
> +1-650-607-0392 <tel:%2B1-650-607-0392> (o)
>
> +1-415-710-8828 <tel:%2B1-415-710-8828> (m)
>
> On 10/16/14, 11:32 AM, Don Seiler wrote:
>
> Yes in 12c DNFS works on NFSv4. In fact, NFSv4 is
> required if you plan to use OISP (Oracle
> Intelligent Storage Protocol) to talk to their
> ZFSSA. Although I would not suggest using the
> ZFSSA to run live databases yet. Should be OK for
> FRA uses.
>
> Don.
>
> On Thu, Oct 16, 2014 at 10:46 AM, Kevin Jernigan
> <kevin.jernigan_at_oracle.com
> <mailto:kevin.jernigan_at_oracle.com>> wrote:
>
> As of Oracle Database 12c, dNFS works with
> both NFSv3 and NFSv4...KJ
>
> --
>
> Kevin Jernigan
>
> Senior Director Product Management
>
> Advanced Compression, Hybrid Columnar
>
> Compression (HCC), Database File System
>
> (DBFS), SecureFiles, Database Smart Flash
>
> Cache, Total Recall, Database Resource
>
> Manager (DBRM), Direct NFS Client (dNFS),
>
> Continuous Query Notification (CQN),
>
> Index Organized Tables (IOT), Information
>
> Lifecycle Management (ILM)
>
> +1-650-607-0392 <tel:%2B1-650-607-0392> (o)
>
> +1-415-710-8828 <tel:%2B1-415-710-8828> (m)
>
> On 10/16/14, 8:37 AM, Hameed, Amir wrote:
>
> I don’t believe DNFS is certified to work
> with NFSv4.
>
> *From:*oracle-l-bounce_at_freelists.org
> <mailto:oracle-l-bounce_at_freelists.org>
> [mailto:oracle-l-bounce_at_freelists.org] *On
> Behalf Of *Hans Forbrich
> *Sent:* Thursday, October 16, 2014 11:30 AM
> *To:* oracle-l_at_freelists.org
> <mailto:oracle-l_at_freelists.org>
> *Subject:* Re: Is filesystemio_options
> relevant when the database is on ASM ?
>
> On 16/10/2014 3:35 AM, Frits Hoogland wrote:
>
> When using NFS underneath ASM, I've
> witnessed filesystemio_options being
> honoured by the database, which means
> it needs setting it to 'setall' for
> the combination AIO+DIO. Which makes
> sense, because you need to create a
> file on a (NFS) filesystem to be used
> as ASM disk device.
>
> Then it becomes important to know which NFS?
>
> I believe DNFS behaves different than
> standard NFSv3 which may be different
> again from NFSv4
>
> /Hans
>
>
>
> --
> Don Seiler
> http://www.seiler.us
>
>
>
> --
> Don Seiler
> http://www.seiler.us
>
>
>
> --
> Don Seiler
> http://www.seiler.us
>
>
>
> --
> Don Seiler
> http://www.seiler.us
>
-- http://www.freelists.org/webpage/oracle-l