From: Jeff Dike <jdike@ka...> - 2002-09-15 12:51:33
> The number of packets is near enough identical, but there's a
> significant difference (~7%) between bandwidth usage of the tap
> device, and the interface on the UML side. Does anyone have an idea
> what could/does account for this?
I wonder if those are dropped packets. 7% seems like a lot, though.
And they might not be showing up as dropped in the stats because I'm not
convinced that the drivers are accounting them properly.
From: David Coulson <david@da...> - 2002-09-15 15:19:03
Jeff Dike wrote:
> I wonder if those are dropped packets. 7% seems like a lot, though.
Indeed. I don't see a drop of 7% in any other interface stats though, so
if it is being dropped, then the packets are being accounted as
> And they might not be showing up as dropped in the stats because I'm not
> convinced that the drivers are accounting them properly.
Well, unless something dropped them after the kernel counted them (since
'packets' is pretty much the same, and certainly not 7% different). As
far as I can see, it's only the actual byte count that is incorrect.
I thought it might be weird stuff like BPDUs from the bridge, but I've
got STP disabled, so I've really got no idea. One of my UMLs has 1.2Mb
on the UML, yet 1.5Mb on the host, even though the packet counts are
near as identical as I can get running them 'ifconfig' commands
seperatly. Certainly, the difference in packet count on the host and the
UML is negligable compared to the change in byte count.
As you said, it could be a UML driver issue. Since the TX bytes on the
UML is equal to the RX on the host, it's sort of confusing, since I
assume it's handled in the same way as packets going t'other way.
David Coulson http://davidcoulson.net/