netdev
[Top] [All Lists]

Re: [PATCH] netem: account for packets in delayed queue in qlen

To: Patrick McHardy <kaber@xxxxxxxxx>
Subject: Re: [PATCH] netem: account for packets in delayed queue in qlen
From: Stephen Hemminger <shemminger@xxxxxxxx>
Date: Tue, 5 Apr 2005 12:58:15 -0700
Cc: "David S. Miller" <davem@xxxxxxxxxxxxx>, netdev@xxxxxxxxxxx
In-reply-to: <4252EB9D.9070305@trash.net>
Organization: Open Source Development Lab
References: <20050329152110.38d50653@dxpl.pdx.osdl.net> <4252EB9D.9070305@trash.net>
Sender: netdev-bounce@xxxxxxxxxxx
On Tue, 05 Apr 2005 21:48:45 +0200
Patrick McHardy <kaber@xxxxxxxxx> wrote:

> Stephen Hemminger wrote:
> > Netem has a private queue for delayed packets, and currently, packets
> > in this queue are not accounted for in the qdisc qlen statistics.
> > This is a problem if netem is used inside another qdisc doing rate
> > control that peeks at the qlen.
> > 
> > This patch changes the statistics to include the packets held but
> > not ready to send.
> 
> There is one troublesome spot left, netem_watchdog() decreases q.qlen
> when the packet couldn't be enqueued. I don't think it is possible
> to make netem useable as leaf-qdisc, it will always have to touch
> q.qlen from timer context and classful qdiscs can't deal with this
> since they all maintain their own q.qlen counters and expect changes
> only in the +-1 range in enqueue/dequeue/requeue/drop. Best thing IMO
> would be to refuse to work as anything but root qdisc.

But then it won't work as a leaf off of prio to handle a single flow.

<Prev in Thread] Current Thread [Next in Thread>