Mailing List Archive

Difference between file and tap:aio
Someone can explain me good difference between file and tap:aio, pros and
cons, and also with all possibility with blktap2? Thanks for any response
--
View this message in context: http://old.nabble.com/Difference-between-file-and-tap%3Aaio-tp28191044p28191044.html
Sent from the Xen - User mailing list archive at Nabble.com.


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users
Re: Difference between file and tap:aio [ In reply to ]
On Fri, Apr 09, 2010 at 07:13:22AM -0700, Fantu wrote:
>
> Someone can explain me good difference between file and tap:aio, pros and
> cons, and also with all possibility with blktap2? Thanks for any response
>

file: uses dom0 kernel page cache, and thus might give better performance than phy: or tap:aio:,
but it's also more insecure because of the caching.

tap:aio: uses direct IO, so it bypasses dom0 kernel caches, and works like phy: in that sense.

blktap1 provides also other modes than :aio, like: qcow, but those were not very stable or fast,
so blktap2 was developed. for blktap1 see: http://wiki.xensource.com/xenwiki/blktap .

blktap2 also has vhd image support including snapshots and cloning.
See here: http://wiki.xensource.com/xenwiki/blktap2 .

-- Pasi


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users
Re: Difference between file and tap:aio [ In reply to ]
Would file: really have a better performance than phy: ? I am planing to
use a own partition for my Windows HVM Domain, cause in the fact you
going to use a Filesystem in a filesystem, so this need more time than
only "one" filesystem i think.....



Am 09.04.2010 16:24, schrieb Pasi Kärkkäinen:
> On Fri, Apr 09, 2010 at 07:13:22AM -0700, Fantu wrote:
>
>> Someone can explain me good difference between file and tap:aio, pros and
>> cons, and also with all possibility with blktap2? Thanks for any response
>>
>>
> file: uses dom0 kernel page cache, and thus might give better performance than phy: or tap:aio:,
> but it's also more insecure because of the caching.
>
> tap:aio: uses direct IO, so it bypasses dom0 kernel caches, and works like phy: in that sense.
>
> blktap1 provides also other modes than :aio, like: qcow, but those were not very stable or fast,
> so blktap2 was developed. for blktap1 see: http://wiki.xensource.com/xenwiki/blktap .
>
> blktap2 also has vhd image support including snapshots and cloning.
> See here: http://wiki.xensource.com/xenwiki/blktap2 .
>
> -- Pasi
>
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@lists.xensource.com
> http://lists.xensource.com/xen-users
>
>


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users
Re: Difference between file and tap:aio [ In reply to ]
On Tue, Apr 13, 2010 at 4:13 AM, Tegger <xen@tegger.de> wrote:
> Would file: really have a better performance than phy: ?

If your working set is smaller than dom0 memory, then yes. This is due
to dom0 caching effect.

> I am planing to use
> a own partition for my Windows HVM Domain, cause in the fact you going to
> use a Filesystem in a filesystem, so this need more time than only "one"
> filesystem i think.....

functionality-wise, there's nothing "wrong" about have a domU
filesystem on top of a disk image file on top of dom0 filesystem.
I'd stick with phy:/ though, due to (generally) more performance and reliablity.

--
Fajar

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users