Does blkback bypass Dom0's memory buffer?

Does blkback bypass Dom0's memory buffer?

[xu cong]

[-- Attachment #1.1: Type: text/plain, Size: 252 bytes --]

Does blkback thread bypass Dom0's memory buffer? If I write in DomU without
O_Direct, the data will be buffered in DomU's memory cache then be flushed
to disk. Will them be buffered in Dom0's memory again? How about the
netback? Thanks.

Regards,
Cong

[-- Attachment #1.2: Type: text/html, Size: 322 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: Does blkback bypass Dom0's memory buffer?

[Wei Liu]

On Tue, Feb 4, 2014 at 9:15 PM, xu cong <congxumail@gmail.com> wrote:
> Does blkback thread bypass Dom0's memory buffer? If I write in DomU without
> O_Direct, the data will be buffered in DomU's memory cache then be flushed
> to disk. Will them be buffered in Dom0's memory again? How about the
> netback? Thanks.
>

Why is netback involved? I don't quite get your question. Block device
has a completely different data path from network.

Wei.

> Regards,
> Cong
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
>

Re: Does blkback bypass Dom0's memory buffer?

[xu cong]

[-- Attachment #1.1: Type: text/plain, Size: 839 bytes --]

Yes, my question is confusing. Please ignore the last sentence. I just
wonder whether block backend thread will use Dom0's memory buffer or not.
Thanks.


2014-02-04 Wei Liu <liuw@liuw.name>:

> On Tue, Feb 4, 2014 at 9:15 PM, xu cong <congxumail@gmail.com> wrote:
> > Does blkback thread bypass Dom0's memory buffer? If I write in DomU
> without
> > O_Direct, the data will be buffered in DomU's memory cache then be
> flushed
> > to disk. Will them be buffered in Dom0's memory again? How about the
> > netback? Thanks.
> >
>
> Why is netback involved? I don't quite get your question. Block device
> has a completely different data path from network.
>
> Wei.
>
> > Regards,
> > Cong
> >
> > _______________________________________________
> > Xen-devel mailing list
> > Xen-devel@lists.xen.org
> > http://lists.xen.org/xen-devel
> >
>

[-- Attachment #1.2: Type: text/html, Size: 1434 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: Does blkback bypass Dom0's memory buffer?

[xu cong]

[-- Attachment #1.1: Type: text/plain, Size: 839 bytes --]

Yes, my question is confusing. Please ignore the last sentence. I just
wonder whether block backend thread will use Dom0's memory buffer or not.
Thanks.


2014-02-04 Wei Liu <liuw@liuw.name>:

> On Tue, Feb 4, 2014 at 9:15 PM, xu cong <congxumail@gmail.com> wrote:
> > Does blkback thread bypass Dom0's memory buffer? If I write in DomU
> without
> > O_Direct, the data will be buffered in DomU's memory cache then be
> flushed
> > to disk. Will them be buffered in Dom0's memory again? How about the
> > netback? Thanks.
> >
>
> Why is netback involved? I don't quite get your question. Block device
> has a completely different data path from network.
>
> Wei.
>
> > Regards,
> > Cong
> >
> > _______________________________________________
> > Xen-devel mailing list
> > Xen-devel@lists.xen.org
> > http://lists.xen.org/xen-devel
> >
>

[-- Attachment #1.2: Type: text/html, Size: 1434 bytes --]

[-- Attachment #2: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: Does blkback bypass Dom0's memory buffer?

[Roger Pau Monné]

On 04/02/14 22:15, xu cong wrote:
> Does blkback thread bypass Dom0's memory buffer? If I write in DomU
> without O_Direct, the data will be buffered in DomU's memory cache then
> be flushed to disk. Will them be buffered in Dom0's memory again? How
> about the netback? Thanks.

Blkback will not do any kind of buffering itself, the request is read
from the shared ring and passed to the underlying device using
submit_bio. If you want to make sure your data has hit the disk you
should issue a flush operation (see "feature-flush-cache" in blkif.h)

Roger.