[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <4D560350.5080002@chelsio.com>
Date: Fri, 11 Feb 2011 19:49:36 -0800
From: Dimitris Michailidis <dm@...lsio.com>
To: Tom Herbert <therbert@...gle.com>
CC: davem@...emloft.net, netdev@...r.kernel.org
Subject: Re: [PATCH 1/2] net: Add RX queue weights
Tom Herbert wrote:
> This patch adds a weight attribute to the netdev RX queues. This allows
> control over the relative receive packet load for each queue. These
> values are set in sysfs variable 'weight' in the rxq directory for
> a device. When a weight is set, a new netdev operation is called to
> inform the driver of the changed weight. The driver is expected to
> apply the queue weights in a logical manner to the RSS indirection table
> of the device to achieve the desired weighting. The driver
> implementation for this is unspecified.
>
> If a weight for a queue is zero, this effectively disables that queue
> for RSS (but possibly still usable by accelerated RFS, etc.), except
> in the case that all queue weights are zero, then all queues are
> considered equally weighted (the default).
>
> Example configuration:
> echo 1 > /sys/class/net/eth4/queues/rx-0/weight
> echo 1 > /sys/class/net/eth4/queues/rx-1/weight
> echo 5 > /sys/class/net/eth4/queues/rx-2/weight
> echo 0 > /sys/class/net/eth4/queues/rx-3/weight
>
> So rx queue 0 and 1 have equal weight, queue 2 is 5X in weight and
> queue 3 is disabled for RSS.
Doesn't ethtool -X already do this? With the added benefit that ethtool
doesn't need each driver to provide its own weight handling arithmetic.
--
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