lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  linux-cve-announce  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20101229084740.GR2754@reaktio.net>
Date:	Wed, 29 Dec 2010 10:47:40 +0200
From:	Pasi Kärkkäinen <pasik@....fi>
To:	"Kirsher, Jeffrey T" <jeffrey.t.kirsher@...el.com>
Cc:	Alkis Georgopoulos <alkisg@...il.com>,
	"netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: Bypass flow control problems

On Tue, Dec 28, 2010 at 09:05:42PM -0800, Kirsher, Jeffrey T wrote:
> On Dec 27, 2010, at 8:43, Pasi Kärkkäinen <pasik@....fi> wrote:
> 
> > On Wed, Dec 22, 2010 at 07:51:31PM +0200, Alkis Georgopoulos wrote:
> >> Hi,
> >> 
> >> I'm an IT teacher/LTSP developer. In LTSP, thin clients are netbooted
> >> from a server and receive a lot of X and remote disk traffic from it.
> >> 
> >> Many installations have a gigabit NIC on the server, an unmanaged
> >> gigabit switch, and 100 Mbps NICs on the clients.
> >> 
> >> With flow control on, the server is limited to sending 100 Mbps to all
> >> the clients. So with 10 thin clients the server can concurrently send
> >> only 10 Mbps to each one of them.
> >> 
> >> On NICs that support it, we turn flow control off and the server can
> >> properly send 100 Mbps to each client, i.e. 1 Gbps to 10 clients.
> >> 
> >> * Is there any way to bypass that problem on NICs that do not support
> >>   turning off flow control, like e.g. realteks?
> >>   I.e. when a client sends a pause signal to the server, instead of the
> >>   server pausing, to continue sending data to another client?
> >>   Or even to limit the amound of data the server sends to each client,
> >>   so that the clients never have to send pause signals?
> >> 
> > 
> > You could set up QoS rules on the server to limit the network speed per client..
> > 
> >> * I really don't understand why flow control is enabled by default on
> >>   NICs and switches. In which case does it help? As far as I
> >>   understand, all it does is ruin gigabit => 100 Mbps connections...
> >> 
> >> * As a side note, since rtl8169 is a very common chipset, is there a
> >>   way to disable flow control for that specific NIC?
> >> 
> >> This problem affects thousands of LTSP installations, we'd much
> >> appreciate your knowledge and feedback on it.
> >> 
> > 
> > Did you try disabling flow control from the switch? 
> 
> He stated the switch was a un-managed switch, so he would be unable to disable flow control on the switch.
> 

Oh, I missed that :) Never mind then.

-- Pasi

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ