Re: 3c59x slow; debug=5 removes the error; why?
Bogdan Costescu <[email protected]> Tue, 18 May 2004 13:02:16 +0200 (CEST)
| Newsgroups | gmane.linux.drivers.vortex.devel |
|---|---|
| Message-ID | <Pine.LNX.4.44.0405181209170.30903-100000@kenzo.iwr.uni-heidelberg.de> |
On Sat, 15 May 2004, Christian Ostheimer wrote: > case 1): insmod 3c59x > 30-80kB/s; 30% overrun errors Given the direction of the transfer, I assume that you are talking about the Rx overruns that ifconfig displays. Rx overruns appear when the host is too slow for the incoming data, which leads to filling up all Rx buffers - at this point the card cannot receive any more packets and everyhting is dropped. This is not a normal situation and based on the description of your notebook it should be powerful enough to handle 100Mbit/s. > case 2): insmod 3c59x debug=5 > 6-10MB/s; 0.02% overruns This is a bit more normal, however the presence of overruns is still not a good sign. Furthermore, increasing "debug" leads to lower performance of the driver as time is spent on the calls to "printk", which should not increase transfer speed... So it's something to do with the timing and coupled with: > PCI: Found IRQ 11 for device 0000:02:00.0 > PCI: Sharing IRQ 11 with 0000:00:1f.1 ... makes me believe that it has something to do with the interrupt sharing. What is the device that is sharing the IRQ with the 3Com NIC? Is is possible to disable this device in BIOS and see if the problem dissapears ? > downloading from notebook to pc and file transfers by scp in both > directions give full speed Hmm, only Rx interrupts seem to be affected... > Comparing tx and rx packets shows that ftp has 0.7MB tx packets with > 33 MB rx and scp has only 0.2 MB tx with 33 MB rx. That's probably because of the TCP retransmissions. -- Bogdan Costescu IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868 E-mail: [email protected] _______________________________________________ vortex mailing list [email protected] http://www.scyld.com/mailman/listinfo/vortex