Re: [PATCH 1/2]: [NET_SCHED]: Make all rate based scheduler work with TSO.

[Bill Fink <billfink@mindspring.com>]

On Wed, 05 Sep 2007, Jesper Dangaard Brouer wrote:

> On Tue, 2007-09-04 at 13:40 -0400, Bill Fink wrote:
> > On Tue, 04 Sep 2007, Patrick McHardy wrote:
> > 
> > > Bill Fink wrote:
> > > > On Sat, 1 Sep 2007, Jesper Dangaard Brouer wrote:
> > > >
> > > Yes, you need to specify the MTU on the command line for
> > > jumbo frames.
> > 
> > Thanks!  Works much better now, although it does slightly exceed
> > the specified rate.
> 
> Thats what happens, with the current rate table system, as we use the
> lower boundry (when doing the packet to time lookups). Especially with a
> high MTU, as the "resolution" of the rate table diminish (mpu=9000 gives
> cell_log=6, 2^6=64 bytes "resolution" buckets).
> 
> > [root@lang4 ~]# tc qdisc add dev eth2 root tbf rate 2gbit buffer 5000000 limit 18000 mtu 9000
> > 
> > [root@lang4 ~]# ./nuttcp-5.5.5 -w10m 192.168.88.14
> >  2465.6729 MB /  10.08 sec = 2051.8241 Mbps 19 %TX 13 %RX

That doesn't seem to account for the magnitude of the rate exceeding.
In the worst case (rough calculation):

	(1+64/9000)*2000 = 2014.2222 Mbps

Now if that were 256 rather than 64:

	(1+256/9000)*2000 = 2056.8888 Mbps

Or maybe the packet overhead is calculated wrong for the 9000 MTU case
(just wild speculation on my part).

						-Bill