poor network performance / TX overruns

poor network performance / TX overruns

Post by Tobias Richt » Wed, 05 Dec 2001 20:51:25



I installed Debian Woody on a Alpha 164 without major problems
execpt the network performance is poor / unreliable.

Using ping -f it loses about 20% packets (to and from the machine).
ifconfig shows:

eth0      Link encap:Ethernet  HWaddr 00:C0:F0:56:CF:D1  
          inet addr:10.17.26.3  Bcast:10.17.26.255  Mask:255.255.255.0
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:8804 errors:1 dropped:0 overruns:0 frame:0
          TX packets:9257 errors:1165 dropped:0 overruns:1164 carrier:1
          collisions:0 txqueuelen:100
          RX bytes:947344 (925.1 KiB)  TX bytes:1035684 (1011.4 KiB)
          Interrupt:18 Base address:0x5000

The system on the other end show no errors/overuns.
The problem looks just the same using either kernel 2.2.20 or 2.4.16
The NIC is a Kingston 110TX using the tulip driver:

        Linux Tulip driver version 0.9.15-pre9 (Nov 6, 2001)
        tulip0:  MII transceiver #1 config 3100 status 7829 advertising 01e1.
        eth0: Lite-On 82c168 PNIC rev 32 at 0xfffffc8602265000, \
                                        00:C0:F0:56:CF:D1, IRQ 18.
        eth0: Setting full-duplex based on MII#1 link partner capability \
                                        of 01e1.

Checking /proc/interrupts and ioports I found on obvious resource
conflict.

Now, what are TX overrruns (I only knew RX overruns)?
What can I do to get rid of these?

TIA,
tobias

 
 
 

poor network performance / TX overruns

Post by Dean Thompso » Wed, 05 Dec 2001 21:45:40


Hi!,


> I installed Debian Woody on a Alpha 164 without major problems
> execpt the network performance is poor / unreliable.

> Using ping -f it loses about 20% packets (to and from the machine).
> ifconfig shows:

> eth0      Link encap:Ethernet  HWaddr 00:C0:F0:56:CF:D1
>           inet addr:10.17.26.3  Bcast:10.17.26.255  Mask:255.255.255.0
>           UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
>           RX packets:8804 errors:1 dropped:0 overruns:0 frame:0
>           TX packets:9257 errors:1165 dropped:0 overruns:1164 carrier:1
>           collisions:0 txqueuelen:100
>           RX bytes:947344 (925.1 KiB)  TX bytes:1035684 (1011.4 KiB)
>           Interrupt:18 Base address:0x5000

> The system on the other end show no errors/overuns.
> The problem looks just the same using either kernel 2.2.20 or 2.4.16
> The NIC is a Kingston 110TX using the tulip driver:

>         Linux Tulip driver version 0.9.15-pre9 (Nov 6, 2001)
>         tulip0:  MII transceiver #1 config 3100 status 7829 advertising 01e1.
>         eth0: Lite-On 82c168 PNIC rev 32 at 0xfffffc8602265000, \
>                                         00:C0:F0:56:CF:D1, IRQ 18.
>         eth0: Setting full-duplex based on MII#1 link partner capability \
>                                         of 01e1.

> Checking /proc/interrupts and ioports I found on obvious resource
> conflict.

> Now, what are TX overrruns (I only knew RX overruns)?
> What can I do to get rid of these?

Just an idea, but have you changed over the cable to make sure that it isn't a
bad fitting.  Additionally, it isn't a case of one end being in full duplex
mode and the other cards being in half-duplex mode is it ?

See ya

Dean Thompson

--
+____________________________+____________________________________________+

| Bach. Computing (Hons)     | ICQ     - 45191180                         |
| PhD Student                | Office  - <Off-Campus>                     |
| School Comp.Sci & Soft.Eng | Phone   - +61 3 9903 2787 (Gen. Office)    |
| MONASH (Caulfield Campus)  | Fax     - +61 3 9903 1077                  |
| Melbourne, Australia       |                                            |
+----------------------------+--------------------------------------------+

 
 
 

poor network performance / TX overruns

Post by Tobias Richt » Wed, 05 Dec 2001 22:19:27




>> Now, what are TX overrruns (I only knew RX overruns)?
>> What can I do to get rid of these?

>Just an idea, but have you changed over the cable to make sure that it isn't a
>bad fitting.  Additionally, it isn't a case of one end being in full duplex
>mode and the other cards being in half-duplex mode is it ?

I (again) checked the cable - no change. Both the alpha and its peers
are connected to a switch in 100MbitFDX mode. I also tried different
ports on the switch.

I also wonder why I always get drop rates of 20% with ping -f

Thanks anyway,
tobias

 
 
 

poor network performance / TX overruns

Post by Karl Heye » Wed, 05 Dec 2001 23:45:14



> I installed Debian Woody on a Alpha 164 without major problems
> execpt the network performance is poor / unreliable.

> Using ping -f it loses about 20% packets (to and from the machine).
> ifconfig shows:

> eth0      Link encap:Ethernet  HWaddr 00:C0:F0:56:CF:D1  
>           inet addr:10.17.26.3  Bcast:10.17.26.255  Mask:255.255.255.0
>           UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
>           RX packets:8804 errors:1 dropped:0 overruns:0 frame:0
>           TX packets:9257 errors:1165 dropped:0 overruns:1164 carrier:1
>           collisions:0 txqueuelen:100
>           RX bytes:947344 (925.1 KiB)  TX bytes:1035684 (1011.4 KiB)
>           Interrupt:18 Base address:0x5000

> The system on the other end show no errors/overuns.
> The problem looks just the same using either kernel 2.2.20 or 2.4.16
> The NIC is a Kingston 110TX using the tulip driver:

>    Linux Tulip driver version 0.9.15-pre9 (Nov 6, 2001)
>    tulip0:  MII transceiver #1 config 3100 status 7829 advertising 01e1.
>    eth0: Lite-On 82c168 PNIC rev 32 at 0xfffffc8602265000, \
>                                    00:C0:F0:56:CF:D1, IRQ 18.
>    eth0: Setting full-duplex based on MII#1 link partner capability \
>                                    of 01e1.

> Checking /proc/interrupts and ioports I found on obvious resource
> conflict.

> Now, what are TX overrruns (I only knew RX overruns)?
> What can I do to get rid of these?

Not 100% on this, not very common, but IIRC it is to do to the
TCP window sizes.  I'm guessing the other machines NIC buffer is
not being emptied quick enough.  What is on the other machine and
what NIC is it using.  Can you xhange the NIC in the other box.

Another possibility is the switch, you could try things like
disabling window scaling.  Maybe try a cross over able as a test

echo 0 > /proc/sys/net/ipv4/tcp_window_scaling

karl.

 
 
 

poor network performance / TX overruns

Post by Tobias Richt » Thu, 06 Dec 2001 00:44:08




>> ifconfig shows:

>> eth0      Link encap:Ethernet  HWaddr 00:C0:F0:56:CF:D1  
>>           RX packets:8804 errors:1 dropped:0 overruns:0 frame:0
>>           TX packets:9257 errors:1165 dropped:0 overruns:1164 carrier:1
>>           collisions:0 txqueuelen:100

>> The system on the other end show no errors/overuns.

>Not 100% on this, not very common, but IIRC it is to do to the
>TCP window sizes.  I'm guessing the other machines NIC buffer is
>not being emptied quick enough.  What is on the other machine and
>what NIC is it using.  Can you xhange the NIC in the other box.

The other boxes NIC should be fine as it works with any other
client without a problem. It's an Intel Linux (2.2.19) system with
a eepro100 NIC. And as stated above ifconfig on the peer
doesn't show errors (almost):

eth0      Link encap:Ethernet  HWaddr 00:02:B3:08:8A:78  
          RX packets:297827 errors:0 dropped:0 overruns:0 frame:0
          TX packets:360181 errors:0 dropped:0 overruns:6 carrier:0
          collisions:0 txqueuelen:100

Slow input handling should give me RX error, I think.

Quote:>Another possibility is the switch, you could try things like
>disabling window scaling.  Maybe try a cross over able as a test

>echo 0 > /proc/sys/net/ipv4/tcp_window_scaling

disabling window scaling didn't help. And I'm afraid the cross over
cable also didn't change anything.

Thanks,
tobias

 
 
 

poor network performance / TX overruns

Post by Tobias Richt » Thu, 06 Dec 2001 01:18:13



>disabling window scaling didn't help. And I'm afraid the cross over
>cable also didn't change anything.

By the way: A month ago the Alpha was running FreeBSD and
networking was fine there. So I'd guess it's not a hardware
issue and not (really) related to its commincation peers.

tobias

 
 
 

poor network performance / TX overruns

Post by Karl Heye » Thu, 06 Dec 2001 01:14:44



>>Another possibility is the switch, you could try things like
>>disabling window scaling.  Maybe try a cross over able as a test

>>echo 0 > /proc/sys/net/ipv4/tcp_window_scaling

> disabling window scaling didn't help. And I'm afraid the cross over
> cable also didn't change anything.

Do you have any messages wrt the network in either dmesg or
/var/log/messages (or whatever your syslog is writing to)

This maybe an issue with the NIC itself, and maybe best to go and
post on the tulip driver mailing list, look on the page
http://www.scyld.com/network/tulip.html for more info.

karl.

 
 
 

poor network performance / TX overruns

Post by Michael Heimin » Thu, 06 Dec 2001 02:05:15



17:18:


>>disabling window scaling didn't help. And I'm afraid the cross over
>>cable also didn't change anything.

> By the way: A month ago the Alpha was running FreeBSD and
> networking was fine there. So I'd guess it's not a hardware
> issue and not (really) related to its commincation peers.

> tobias

Some alpha machines do have problems with the speed settings of the
NIC, did you try checking/setting the speed on the SRM console
(presuming you boot of SRM)

Quote:>>> show ewa0_mode
>>> set ewa0_mode FastFD

(Assuming you NIC is ewa0, this may be different depending on the SRM
console version you run)

You may find additional info at gatekeeper.dec.com or
www.linuxalpha.org (There is a mailinglist).

Just out of interest, is there a command similar to 'consvar' (Tru64)
to control SRM console settings, if you run Linux on alpha?

Good luck

Michael Heiming

 
 
 

poor network performance / TX overruns

Post by Tobias Richt » Thu, 06 Dec 2001 02:58:26



>Some alpha machines do have problems with the speed settings of the
>NIC, did you try checking/setting the speed on the SRM console
>(presuming you boot of SRM)

>>>> show ewa0_mode
>>>> set ewa0_mode FastFD

>(Assuming you NIC is ewa0, this may be different depending on the SRM
>console version you run)

As far as I can tell SRM does not know anything about my
NIC at all. "set ewa0_mode FastFD" created a new variable
but didn't help.

Quote:>Just out of interest, is there a command similar to 'consvar' (Tru64)
>to control SRM console settings, if you run Linux on alpha?

I guess it prints out SRM's variables. Not that I've heard
of or seen.

Thanks,
tobias

 
 
 

poor network performance / TX overruns

Post by Tobias Richt » Thu, 06 Dec 2001 03:09:18



>This maybe an issue with the NIC itself, and maybe best to go and
>post on the tulip driver mailing list, look on the page
>http://www.scyld.com/network/tulip.html for more info.

The problem is definately in the driver or NIC. I've now replaced
both with 8139too models and everything started working fine.

I'll just drop the Kingston NIC into my 'bogus hardware' box.

Many thanks,
tobias

 
 
 

poor network performance / TX overruns

Post by Michael Heimin » Thu, 06 Dec 2001 03:17:38



18:58:


>>Some alpha machines do have problems with the speed settings of the
>>NIC, did you try checking/setting the speed on the SRM console
>>(presuming you boot of SRM)

>>>>> show ewa0_mode
>>>>> set ewa0_mode FastFD

>>(Assuming you NIC is ewa0, this may be different depending on the
>>SRM console version you run)

> As far as I can tell SRM does not know anything about my
> NIC at all. "set ewa0_mode FastFD" created a new variable
> but didn't help.

No, stop you shouldn't try this, it was just an example, how on earth
should I know what NIC is in your box?

Try:

Quote:>>> sho dev
>>> show ew*mode

Try setting something not possible, for the card the above command
shows:

Quote:>>> set <ew*> fadfdfg

The system should now answer with the settings possible.
Didn't you got any manual with your box? This should be documented.

Quote:

>>Just out of interest, is there a command similar to 'consvar'
>>(Tru64) to control SRM console settings, if you run Linux on alpha?

> I guess it prints out SRM's variables. Not that I've heard
> of or seen.

No, you can although set parameters with 'consvar' to the nvram, not
only show.

Michael Heiming

 
 
 

poor network performance / TX overruns

Post by M. Buchenriede » Thu, 06 Dec 2001 01:12:44


[Note FollowUp-To: header]

[...]

Quote:>eth0      Link encap:Ethernet  HWaddr 00:C0:F0:56:CF:D1  
>          inet addr:10.17.26.3  Bcast:10.17.26.255  Mask:255.255.255.0
>          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
>          RX packets:8804 errors:1 dropped:0 overruns:0 frame:0
>          TX packets:9257 errors:1165 dropped:0 overruns:1164 carrier:1
>          collisions:0 txqueuelen:100
>          RX bytes:947344 (925.1 KiB)  TX bytes:1035684 (1011.4 KiB)
>          Interrupt:18 Base address:0x5000

[...]

Your NIC is in full-duplex mode while the link partner is in half-duplex
mode. If that's a hub, you must force half-duplex on the NIC, as a hub
doesn't support FD connections.

Michael
--

          Lumber Cartel Unit #456 (TINLC) & Official Netscum
    Note: If you want me to send you email, don't munge your address.

 
 
 

poor network performance / TX overruns

Post by Tobias Richt » Thu, 06 Dec 2001 07:23:56



>No, stop you shouldn't try this, it was just an example, how on earth
>should I know what NIC is in your box?

>Try:
>>>> sho dev
>>>> show ew*mode

Ah, ok.

Quote:>Try setting something not possible, for the card the above command
>shows:

>>>> set <ew*> fadfdfg

>The system should now answer with the settings possible.
>Didn't you got any manual with your box? This should be documented.

It's a second (or third) hand system. It came with something
else installed. So I downloaded and updated the firmware to SRM
myself. I didn't bother reading the online resources after
I was able to boot the thing into Linux.
I'll have a look at it again, now I know that there are useful
configuration options hidden.

Thanks again,
tobias

 
 
 

1. sl0 RX overrun 20 TX overrun 57

Hi all!

Any comment about the following?

I have two Linux boxes connected via CSLIP and direct wire.
No matter whether the speed is 9600, 19200 or 38400 bauds.
Kernel version is 1.1.37 (but 1.0.8 acts in a similar way).

Routing is set up correctly, ping works brilliantly in both
directions. Telnet works so slowly that it times out before
I can log in. Ftp also works very slowly. ps shows that
in.telnetd on the server side is running. Ifconfig shows
that some overrun of packets occurs on sl0 (ethernet works OK):

----------------------------------------------------------
pygalik:/etc/rc.d# ifconfig
lo        Link encap Local Loopback
inet addr 127.0.0.1  Bcast 127.255.255.255  Mask 255.0.0.0
UP BROADCAST LOOPBACK RUNNING  MTU 2000  Metric 1
RX packets 0 errors 0 dropped 0 overrun 0
TX packets 52 errors 0 dropped 0 overrun 0

sl0       Link encap Serial Line IP
inet addr 200.40.40.95  P-t-P 200.40.40.96  Mask 255.255.255.0
UP POINTOPOINT RUNNING  MTU 296  Metric 1
RX packets 3058 errors 0 dropped 0 overrun 20
TX packets 3293 errors 0 dropped 0 overrun 57

eth0      Link encap 10Mbps Ethernet  HWaddr 00:00:E8:C3:D4:97
inet addr 193.40.127.4  Bcast 193.40.127.255  Mask 255.255.255.0
UP BROADCAST RUNNING  MTU 1500  Metric 1
RX packets 4345 errors 0 dropped 0 overrun 0
TX packets 4987 errors 0 dropped 0 overrun 0

pygalik:/etc/rc.d# netstat -i
Kernel Interface table
Iface   MTU Met  RX-OK RX-ERR RX-DRP RX-OVR  TX-OK TX-ERR TX-DRP TX-OVR Flags
lo     2000   0      0      0      0      0     52      0      0      0 BLRU
sl0     296   0   3066      0      0     21   3308      0      0     63 PRU
eth0   1500   0   5045      0      0      0   5919      0      0      0 BRU
pygalik:/etc/rc.d#
---------------------------------------------------------------------------
Can anyone explain, what does this overrun mean
and what influence has it on performance?

                                        Anto,

2. need a kmail extraction script

3. Poor 100Base-TX Performance

4. IPFWADM Question

5. Poor networking performance

6. how can I send a sound?

7. Poor performance on the network between Sol8 and redhat 7.3 and 7.2

8. Serving Database via Linux Apache??

9. PROBLEM: high system usage / poor SMP network performance

10. Poor Python/Perl network performance under Solaris

11. Poor network performance?

12. Poor network performance after installing AGP board

13. PROBLEM: high system usage / poor SMP network performance