SCIENTIFIC-LINUX-USERS Archives

May 2005

SCIENTIFIC-LINUX-USERS@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Michael Mansour <[log in to unmask]>
Reply To:
Michael Mansour <[log in to unmask]>
Date:
Sun, 22 May 2005 11:00:25 +1000
Content-Type:
text/plain
Parts/Attachments:
text/plain (198 lines)
Hi Michel,

I know it's been a while since we last chatted about this, but I took your 
recommendation at that time and bought 4 of these cards from the US (I'm in 
Australia).

When inserting the cards into 2 newly built SL4 servers, Linux picks them up 
and configures the e1000 module without issues.

lspci shows:

00:0e.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet Controller 
(rev 02)

and 

00:09.0 Ethernet controller: Intel Corp. 82540EM Gigabit Ethernet Controller 
(rev 02)

on both servers, and ethtool shows:

Settings for eth1:
        Supported ports: [ TP ]
        Supported link modes:   10baseT/Half 10baseT/Full
                                100baseT/Half 100baseT/Full
                                1000baseT/Full
        Supports auto-negotiation: Yes
        Advertised link modes:  10baseT/Half 10baseT/Full
                                100baseT/Half 100baseT/Full
                                1000baseT/Full
        Advertised auto-negotiation: Yes
        Speed: 1000Mb/s
        Duplex: Full
        Port: Twisted Pair
        PHYAD: 0
        Transceiver: internal
        Auto-negotiation: on
        Supports Wake-on: umbg
        Wake-on: g
        Current message level: 0x00000007 (7)
        Link detected: yes

on both servers.

However, I've been testing "ping" on these cards under SL4 without any 
success (a ping from the local host works but not from the remote host).

I inserted one card in each of two servers, and hooked them up via a cross-
over cable. They autonegotiate to 1000mbps FD, but they cannot communicate 
with each other at all (they link fine, just cannot ping each other). When I 
try adn ping from the private subnet I've assigned each card, the only 
responses I get from the cards are:

kernel: NETDEV WATCHDOG: eth1: transmit timed out
kernel: e1000: eth1: e1000_watchdog: NIC Link is Down
kernel: e1000: eth1: e1000_watchdog: NIC Link is Up 1000 Mbps Full Duplex

every few seconds. If I stop the pings, they sit there idle and happy 
without resetting the interfaces at all, only when I try and transmit data 
(ping), I get the "NETDEV WATCHDOG" errors.

Last night I was trouble-shooting this problem for about 6 hours, trying 
everything I could think of, different speeds and duplex settings, even 
downloading the latest intel drivers for these cards and compiling them as 
modules and trying them, still no go (although the new driver - e1000-
6.0.54.tar.gz - got rid of the "NETDEV WATCHDOG" errors, I still couldn't 
ping the interfaces).

I used the intel diag1000 tools off the cd which accompanied the cards to 
test the cards from dos. All test passed fine and even the continuous 
network tests with one setup as a responder and the other running the 
network test (and vice versa), but under Linux a total no go.

When running ethtool's offline test, on one server I get:

# ethtool -t eth1 offline
The test result is FAIL
The test extra info:
Register test  (offline)         0
Eeprom test    (offline)         0
Interrupt test (offline)         4
Loopback test  (offline)         0
Link test   (on/offline)         0

(an SMP server)

on the other server I get:

# ethtool -t eth1 offline
The test result is PASS
The test extra info:
Register test  (offline)         0
Eeprom test    (offline)         0
Interrupt test (offline)         0
Loopback test  (offline)         0
Link test   (on/offline)         0

(a UP server)

The "Interrupt test (offline)         4" on the first server concerned me 
but I tried everything I could think of from BIOS changes etc but could not 
get it corrected.

I booted both servers with "noapic" and "acpi=off" as some others suggested 
from googling, but the same results above.

I tried configuring the "InterruptThrottleRate" in /etc/modprobe.conf as 
suggested by Intel in:

ftp://aiedownload.intel.com/df-support/2897/ENG/README.txt

as they seemed to recognise this NETDEV problem, but again no go.

So basically, from what I've seen so far, I just can't use these cards under 
SL4.

Today I'm just going to try connecting them to my 100Mbps FD switch and see 
if they at least work like that, but I'm really not that hopeful.

Michael.

> If you want 32-bit pci Nic's, I'd suggest the Intel Pro/1000 MT.
> 
> It's what I run in this particular desktop system as well as several
> others. In my servers (~50) I have all 64-bit PCI-X broadcom nic's
> integrated on the boards.
> 
> They both seem to work quite well.
> 
> Hope this helps.
> 
> Michael
> 
> On Tue, 2005-02-15 at 10:19 +1000, Michael Mansour wrote:
> > Hi Michael,
> > 
> > Is there any particular model of Intel gigabit card you recommend?
> > 
> > I've spoken to my supplier and they say they'll take back the cards I 
bought 
> > and swap them for Intel, but would like to know a model so I don't have 
to 
> > make this same mistake again.
> > 
> > Thanks.
> > 
> > Michael.
> > 
> > > Well the first thing I'd say is that the realtek chipset's are terrible
> > > if you really want performance. They don't do any packet header
> > > processing or other tcp hardware offloading, causing your CPU to take
> > > the brunt of the packet load (with gigabit this is bad).
> > > 
> > > I would recommend Broadcom or Intel based gigabit cards.
> > > 
> > > Regardless of this fact, mii-tool doesn't support reading out gigabit
> > > link status. It'll give you a link up status with 100FD for 1000FD 
cards
> > > linked at 1000FD with flow control enabled.
> > > 
> > > I know it at least reports on the broadcom and intel based cards 
> > > with a link up status.
> > > 
> > > Otherwise, I think you might need to find another tool to get real link
> > > status out of a gigabit nic.
> > > 
> > > On Tue, 2005-02-15 at 09:32 +1000, Michael Mansour wrote:
> > > > Hi,
> > > > 
> > > > I've just purchased some Netgear 1gigabit ethernet cards and to my 
> > > > dissapointment mii-tool couldn't be used to query them, even though 
the 
> > linux 
> > > > kernel has no issues with using them.
> > > > 
> > > > I run cluster software with SL303 which uses mii-tool to do link 
level 
> > > > checking etc... so my question is, which 1gigabit cards work 
correctly 
> > with 
> > > > SL303 and mii-tool? so that I get output similar to the following:
> > > > 
> > > > [root@anaconda root]# mii-tool
> > > > eth0: negotiated 100baseTx-HD, link ok
> > > > eth1: negotiated 100baseTx-HD, link ok
> > > > SIOCGMIIPHY on 'eth2' failed: Operation not supported
> > > > 
> > > > note: eth2 is the Netgear card (which uses a Realtek chip), the 
other two 
> > are 
> > > > just standard Realtek PCI cards.
> > > > 
> > > > Thanks.
> > > > 
> > > > Michael.
> > ------- End of Original Message -------
------- End of Original Message -------

ATOM RSS1 RSS2