Re: [SPDK] Problem with Blobstore when write 65MB continously

[Andrey Kuzmin <andrey.v.kuzmin at gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 2417 bytes --]

On Wed, Jan 10, 2018, 19:47 Luse, Paul E <paul.e.luse(a)intel.com> wrote:

> Damn, I should have known that :( Thanks guys!!
>
> So I'll take the easy way out and just ask.. what's the most efficient way
> for an app, designed just this one for whatever reason, essentially
> relinquish control to the event reactor - is there a single call he can
> slip into the loop to give other pending events, if any a chance to run and
> if not continue on submitting?
>

It appears quite logical to start submission with a check for pending
completions, doesn't it? Or check for completions if downstream bdev
returns busy status. That would definitely meet app expectations whatever
the request pool size is.

Regards,
Andrey


> Thx
> Paul
>
> -----Original Message-----
> From: SPDK [mailto:spdk-bounces(a)lists.01.org] On Behalf Of Walker,
> Benjamin
> Sent: Wednesday, January 10, 2018 9:33 AM
> To: freeman.zhang1992(a)gmail.com; spdk(a)lists.01.org
> Subject: Re: [SPDK] Problem with Blobstore when write 65MB continously
>
> On Wed, 2018-01-10 at 16:21 +0000, Harris, James R wrote:
> > Hi Paul and Zhengyu,
> >
> > The problem is that the app is not giving the block device a chance to
> > complete any I/O while submitting the 520 back-to-back requests.
> > Blobstore is passive here – it does not do any polling on the block
> > device – that is up to the application.
>
> To additionally clarify - the bdev layer will poll for completions on your
> behalf, but it does so on the same thread that you are submitting I/O from.
> If you are in a tight loop submitting I/O and never yield back to the event
> reactor, the polling won't have a chance to occur. You can either increase
> the number of reqs per channel or submit smaller batches at a time.
> Basically, do what Jim said below.
>
> > Increasing the number of channel reqs would work – but at some point
> > these will still run out.  So it really depends on your application –
> > either increase the channel reqs to the absolutely maximum you will
> > ever need, or add ENOMEM handling.
> _______________________________________________
> SPDK mailing list
> SPDK(a)lists.01.org
> https://lists.01.org/mailman/listinfo/spdk
> _______________________________________________
> SPDK mailing list
> SPDK(a)lists.01.org
> https://lists.01.org/mailman/listinfo/spdk
>
-- 

Regards,
Andrey

[-- Attachment #2: attachment.html --]
[-- Type: text/html, Size: 3473 bytes --]