[CentOS] Ethernet poor performance

Robert Moskowitz rgm at htt-consult.com
Wed Jul 2 17:37:12 UTC 2008


Warren Young wrote:
> Robert Moskowitz wrote:
>>
>> I get pings around 60ms.
>
> Pings within the same LAN? If so, that's slow even for 100BaseT. It 
> should be under 10 ms.

Well, perhaps I did not test everything out with the good card. MIght 
have been doing only 2 hop tests. But with the bad one, it is bad.

>
>> When I switch the cards around, the addon card attached to my
>> network, I get pings that alternate with one being ~1488ms and the
>> next 488ms! This is regular.
>
> Have you removed the cabling from consideration? And, do you know you 
> have good cabling, tested as following the wiring standards? Hand-made 
> cables are particularly suspect, if made by someone who doesn't know 
> what TIA/EIA 568 is, for example.
Some cables are hand made, some purchased. All tested with my cable 
tester. I know all too well the various cabling standards.
>
>> I have played around with the IRQ for the 2nd card. The 1st card I
>> tried I was limited to 9,10, or 11. The card I am using now allows
>> 3, 4, or 7 (btw, I have disabled serial and parallel ports).
>
> None of this should matter in the PCI world, unless you're using 
> really old hardware, which didn't properly support IRQ sharing.

It is old hardware. Compaq SFFs going back a number of years.

>
>>> - What driver is it using?
>>
>> How do I tell?
>
> grep eth0 /etc/modprobe.conf

Good card: alias eth0 e100

Bad card: alias eth1 8139too

>
> or scan through the output of the dmesg command, looking for stuff 
> that appears near "eth0".

e100: Intel(R) PRO/100 Network Driver, 3.5.10-k2-NAPI
e100: Copyright(c) 1999-2005 Intel Corporation
ACPI: PCI Interrupt 0000:00:0a.0[A] -> Link [LNKD] -> GSI 10 (level, 
low) -> IRQ 10
e100: eth0: e100_probe: addr 0x41200000, irq 10, MAC addr 00:50:8B:00:43:B3
8139cp: 10/100 PCI Ethernet driver v1.2 (Mar 22, 2004)
8139cp 0000:00:0e.0: This (id 10ec:8139 rev 10) is not an 8139C+ 
compatible chip
8139cp 0000:00:0e.0: Try the "8139too" driver instead.
input: PC Speaker as /class/input/input2
8139too Fast Ethernet driver 0.9.27
ACPI: PCI Interrupt 0000:00:0e.0[A] -> Link [LNKD] -> GSI 10 (level, 
low) -> IRQ 10
eth1: RealTek RTL8139 at 0xd08fa000, 00:50:ba:42:82:49, IRQ 10
eth1: Identified 8139 chip type 'RTL-8139B'

The IRQs above are overrides on the BIOS settings of 11 and 3 respectively.

>
>>> - Can you verify that the speed and duplex settings match on both 
>>> ends of
>>> the connection?
>>
>> The switch has its 100Mb LED on. One of the switch ports has my 
>> Speedstream router which is only 10Mb, so we can believe the 100Mb 
>> LED. This is a dumb switch (my public network, so I am not going to 
>> plug into one of my Procurves).
>
> That covers speed, but doesn't say anything about the duplex setting. 
> You should see this mentioned near the Ethernet driver load lines in 
> dmesg, too.

Do you see anything in the lines I pasted above? Those are the only ones 
from dmesg.

>
>> ho, ho. MTU of 1500. Is this not doing MTU path discovery?
>
> PMTU-D is done per-connection at the TCP layer, not statically down at 
> L2. 

That I know, just thought there would be some indication in ifconfig. 
But thinking about it, I am not supprised there is not.





More information about the CentOS mailing list