Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 10 Jul 2014 23:50:03 +0000
From:      "Bjoern A. Zeeb" <bzeeb-lists@lists.zabbadoz.net>
To:        Clint Armstrong <clint@clintarmstrong.net>
Cc:        freebsd-jail@freebsd.org
Subject:   Re: VNET performance
Message-ID:  <221F8CBD-0763-4457-A587-948E887FAD17@lists.zabbadoz.net>
In-Reply-To: <CAJMTyCF956CYHa6JVQ7zuHODM5im%2B3%2B4MPmSTnqDCy7HXiSD8g@mail.gmail.com>
References:  <CAJMTyCF956CYHa6JVQ7zuHODM5im%2B3%2B4MPmSTnqDCy7HXiSD8g@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help

On 10 Jul 2014, at 20:45 , Clint Armstrong <clint@clintarmstrong.net> =
wrote:

> What is the expected network performance of a VNET jail for network
> communication between the jail and the host, or between multiple =
jails? I
> expected it to approach the 10Gbps of the epair device, but I'm not =
seeing
> that.
>=20
> I see between 800 - 1200 Mbps in standard iperf tests both between the =
host
> bridge interface and the vnet jail inteface. I see the same poor =
speeds if
> I make 2 vnet jails and put one side of the epair in each and test =
between
> them.
>=20
> Is the overhead of vnet causing this? Is there anything I can do to =
improve
> this performance.
>=20
> I=92ve tested and seen similar performance on 10.0-RELEASE and =
11.0-CURRENT.

epair has a netisr queuing in between as you cannot call the input =
routines directly from the output routines.  I was able to get a bit =
more traffic through by doing pinning games.

I wonder what a vale switch for vnets could achieve.

=97=20
Bjoern A. Zeeb             "Come on. Learn, goddamn it.", WarGames, 1983




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?221F8CBD-0763-4457-A587-948E887FAD17>