ignite-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Raymond Wilson <raymond_wil...@trimble.com>
Subject RE: Ignite Direct I/O plugin description added to wiki
Date Tue, 20 Mar 2018 23:18:47 GMT
Denis,

Thanks for the clarification on read-intensive workloads not being suitable
for Direct IO.

As a counterpoint to that, I would have thought the working set represented
by the in-memory data held by the Ignite node would be significantly larger
than the OS page cache which should mean the read hit rate on the OS page
cache would be very low under those workloads? Of course, this assumes your
WAL page size is configured to be similar to the OS page size, but still....

-----Original Message-----
From: Denis Magda [mailto:dmagda@apache.org]
Sent: Wednesday, March 21, 2018 11:59 AM
To: dev@ignite.apache.org; Prachi Garg <pgarg@gridgain.com>
Subject: Re: Ignite Direct I/O plugin description added to wiki

*Dmitriy*, thanks. Astonishing job! We'll add a section to the durable
memory tuning page and refer to the wiki for more details:
https://issues.apache.org/jira/browse/IGNITE-7466

Please clarify the following:

> Direct I/O mode can't be enabled for Write Ahead Log files. However,
> when working with plugin, WAL manager applies advising Linux systems
> do not store the data of the file in page cache as they are not required.


For me, it means that WAL always goes through the operating system I/O
calls. Nothing changes for the WAL. However, I'm not sure what you meant to
explain by saying "when working with the plugin (Direct I/O) WAL manager
applies...". Could you rephrase it to bring more clarity?

*Raymond,*

If Direct I/O is enabled by default it will bring down the performance of
read-intensive application because, as Dmitry says, the reads bypass page
cache. So, I would recommend using it for write-intensive workloads and,
probably, for mixed-workloads depending on the reads and writes rate.

--
Denis


On Tue, Mar 20, 2018 at 2:29 PM, Raymond Wilson <raymond_wilson@trimble.com>
wrote:

> Looks good!
>
> Is there any reason why this should not be a default setting if it
> gracefully downgrades to non-Direct IO if not supported by the OS?
>
> Thanks,
> Raymond.
>
> -----Original Message-----
> From: Dmitriy Setrakyan [mailto:dsetrakyan@apache.org]
> Sent: Wednesday, March 21, 2018 10:23 AM
> To: dev <dev@ignite.apache.org>
> Subject: Re: Ignite Direct I/O plugin description added to wiki
>
> Thanks Dmitry, awesome work!
>
> On Wed, Mar 21, 2018 at 12:21 AM, Dmitry Pavlov
> <dpavlov.spb@gmail.com>
> wrote:
>
> > Hi Igniters,
> >
> > I've added description of new plugin for Direct I/O for native
> > persistence (
> > https://issues.apache.org/jira/browse/IGNITE-6341)  to wiki
> > https://cwiki.apache.org/confluence/display/IGNITE/
> > Ignite+Persistent+Store+-+under+the+hood#IgnitePersistentStore-
> > underthehood-DirectI/O
> >
> >
> > SIncerely,
> > Dmitriy Pavlov
> >
>

Mime
View raw message