Re: network problem
Andi Kleen (ak@muc.de)
Tue, 21 Sep 1999 11:37:50 +0200
On Mon, Sep 20, 1999 at 09:27:11PM +0200, Henner Eisen wrote:
> Hi,
>
> >>>>> "Andi" == Andi Kleen <ak@muc.de> writes:
>
> Andi> That does not change anything if packets from the old scp
> Andi> are still in the device queue. Only ifconfig down/up clears
> Andi> it.
>
> >> >From looking at net/sched/sch_generic.c, this is strange as
> >> the generic (default) packet scheduler always refers directly
> >> to the net_device´s tx_queue_len variable. (only the fifo
> >> scheduler maintains a copy of tx_queue_len, which would explain
> >> this, but that scheduler is not the default one).
>
> Andi> Why is this strange?
>
> Maybe I did not make describe the scenario clearly enough. I meant it
> would be strange if
>
> - device uses the default (sch_generic) packet scheduler
> - tx_queue_len is ifconfig´ed to a very low value while the device
> queue is empty
> - after that a large transfer is started
> - during this transfer netstat show large amounts of data in the queues.
There are more queues than just the device queue, e.g. TCP's write/retransmit
queue. If the application uses big writes that is perfectly normal. TCP
keeps the packets because it is waiting in slow start for the other end
to open the window. If you want to measure the qdisc qlen you have to use
tc qdisc with rtnetlink.
-Andi
--
This is like TV. I don't like TV.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/