Re: Linux fs.aio-max-nr Leak?
Date: Mon, 29 Sep 2014 15:23:21 -0400
Message-ID: <CAEidWqMd=K6RL+ToiHihQq4RWzCSa+x054PFq9SKVdfyR9qHug_at_mail.gmail.com>
BTW, if anyone is interested in how to see the current and max settings. The max is set in /etc/sysctl.conf.
cat /proc/sys/fs/aio-max-nr
cat /proc/sys/fs/aio-nr
On Mon, Sep 29, 2014 at 12:46 PM, Kenny Payton <k3nnyp_at_gmail.com> wrote:
> I'm not looking to start another flash debate but I will say we have found
> it to be very reliable.
>
> This particular problem is not limited to environments with flash storage
> and I have observed the behavior across numerous environments for a number
> of years. It certainly seems like a leak and the graphed trend shows the
> same. The most recent occurrence was after 165 days of being up on a very
> active database.
>
> The fact that Oracle's pre-install rpm sets this to 1M leads me to believe
> they are aware of the issue. 1M outstanding aio events seem a bit
> excessive even for a busy database. If the requests were being managed
> properly I would suspect we would see a significant drop off during off
> hours but the trend is to vary slightly throughout the day but day to day
> the trend line continues to climb.
>
>
> I just came across this Pythian article where they did more digging than I
> back in 2012. They claim Oracle support recommending setting aio-max-nr to
> 50M.
>
> http://www.pythian.com/blog/troubleshooting-ora-27090-async-io-errors/
>
>
>
> On Mon, Sep 29, 2014 at 12:09 PM, Alessandro Vercelli <alever22_at_zoho.com>
> wrote:
>
>> Maybe the debate abount pros and cons of flash/SSD storage vs magnetic
>> hard drives has been taken thousands of times without a winner, but I
>> personally would use a non-magnetic disk only for os filesystems (and being
>> compelled to use flash storage), first of all because of performance
>> degradation.
>> In the specific situation, I'm almost certain the problem is due to flash
>> storage.
>>
>> Greetings,
>> Alessandro
>>
>>
>> ---- On Mon, 29 Sep 2014 16:50:46 +0200 Kenny Payton wrote ----
>>
>> >
>> >These are pretty active databases. The instance in question for this
>> event is 20T, all flash storage, in size and runs around 15k iops. Oracle
>> Linux 6.3.
>> >
>> >
>>
>> --
>> http://www.freelists.org/webpage/oracle-l
>>
>>
>>
>
-- http://www.freelists.org/webpage/oracle-lReceived on Mon Sep 29 2014 - 21:23:21 CEST