[Click] problems with Intel e1000GT driver and strange tests behaviour

[Sacchi Alessio]

Hi,

thanks again to Beyers for suggestions about the e1000 driver.

Our e1000GT based on 82541PI (model: PWLA8391GT) works fine with e1000.o
compiled from the given CD, but the e1000.o obtained compiling the click CVS
driver gives this error:

e1000.o: e1000.o: unresolved symbol local_save_flags

so we are not able to poll the devices.

I also tried the trick explained in one of the readme files about this error:
(cd /usr/src/linux; make cloneconfig; make dep) but this did not solved the
matter on my Debian 2.4.27.

Anyone else faced this error?

I would also ask your thinking about the test we made yesterday.

We have a click router switching between a couple of e1000 (not yet pollable), a
RUDE UDP sender (PC A) on another PC equipped with an e1000 and a Gigabit laptop
receiver (PC B).

We made a couple of tests (NICs are negotiating 1Gbit/s)

Test 1: Sending UDP packets at encreasing rates to find the packet loss
treshold. This accours for rates > 1 Mbit/s on wire.

Test 2: FTP session for transferring a file form A to B (same direction of Test
1 for data). The TCP protocol is able to estabilish a 28 Mbit/s thanks to his
algorithms and never gives a retransmission request.

These results are quite strange because router queues never gets full, the
click-router forwards every packet it gets and the gap between udp and tcp is
incredible.

Our target is to track a kind of loss free Throughput/Packet length plot, but
our plots are really far from what we found in licterature.

Every suggestion is welcome.

Alessio.



----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.