Terrible 100Mbit Ethernet performance with 3Com Fast Etherlink XL

Terrible 100Mbit Ethernet performance with 3Com Fast Etherlink XL

Post by Tomas Stephans » Thu, 24 Jul 1997 04:00:00



Hi
Hi have a Dell OptiPlex GXPro 200MHz running Solaris x86 2.5 update 10
we have added a 100Mbit Ethernet card (3Com Fast Etherlink XL) and
was hoping for some performance but it turned out that it was 100 slower
than 10Mbit ethernet on receiving on transmitting it was acceptable
performance.

When running ttcp between a Sun Ultra-170E and the PC we get
Sun 170E    -> PC 100Mbit    94KB/s ( 0.752Mbit/s)
PC 100Mbit  -> Sun 170E         3276KB/s (26Mbit/s)
Sun 170E    -> PC  10Mbit*  1092KB/s ( 8.7Mbit/s)
Sun 170E    -> Sun SS20    2730KB/s (21Mbit/s)

*another GXPro with default builtin Ethernet card 10Mbit.

snoop shows that there seems to be a 0.2s delay ever so often (surfers is the PC and
sunset is the Sun)

Any suggestions? is it a bug in the Etherlink III driver or is the card not upp to
standard or what?

An observation is that when you move the cable from a 10/100Mbit slot
in a switch to a 10 or 100Mbit fixed slot the whole machine just hangs I guess that
has to do with full duplex i.e. moving it from a switch that handles full duplex to
one that does not.

Regards Tomas

Bootup messages
PCI-device: pci1011,210
Ethernet address = 0:c0:4f:c2:d7:8
elx0 (@0x0): 3COM EtherLink III: ether (twpair) 0:c0:4f:c2:d7:8
PCI-device: pci10b7,590032
pci10b7,590032 is /pci@0,0/pci1011,21@e/pci10b7,5900@8
elx1 (@0x0): 3COM EtherLink III: ether (unknown) 0:60:97:bc:57:49
PCI-device: pci10b7,905034
pci10b7,905034 is /pci@0,0/pci10b7,9050@11
(There is an built in 10Mbit card that is enabled (tried also disabling it)

Snoop trace
Using device /dev/elx (promiscuous mode)
  0.00000       sunset -> surfers      TCP D=5001 S=34018 Syn Seq=2341446512 Len=0 Win=8760
  0.00036      surfers -> sunset       TCP D=34018 S=5001 Syn Ack=2341446513 Seq=1073031521 Len=0 Win=8760
  0.00014       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341446513 Len=0 Win=8760
  0.00282       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341446513 Len=1460 Win=8760
  0.00017      surfers -> sunset       TCP D=34018 S=5001     Ack=2341447973 Seq=1073031522 Len=0 Win=8760
  0.00045       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341447973 Len=1460 Win=8760
  0.00026       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341449433 Len=1460 Win=8760
  0.00018      surfers -> sunset       TCP D=34018 S=5001     Ack=2341450893 Seq=1073031522 Len=0 Win=8760
  0.00043       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341450893 Len=1460 Win=8760
  0.00026       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341452353 Len=1460 Win=8760
  0.00018       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341453813 Len=892 Win=8760
  0.04818      surfers -> sunset       TCP D=34018 S=5001     Ack=2341454705 Seq=1073031522 Len=0 Win=8760
  0.00041       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341454705 Len=1460 Win=8760
  0.00029       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341456165 Len=1460 Win=8760
  0.00025       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341457625 Len=1460 Win=8760
  0.00016      surfers -> sunset       TCP D=34018 S=5001     Ack=2341456165 Seq=1073031522 Len=0 Win=8760
  0.00005      surfers -> sunset       TCP D=34018 S=5001     Ack=2341459085 Seq=1073031522 Len=0 Win=8760
  0.00036       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341460545 Len=1460 Win=8760
  0.00021       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341462005 Len=892 Win=8760
  0.00026       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341462897 Len=1460 Win=8760
  0.00012      surfers -> sunset       TCP D=34018 S=5001     Ack=2341459085 Seq=1073031522 Len=0 Win=8760
  0.00006      surfers -> sunset       TCP D=34018 S=5001     Ack=2341459085 Seq=1073031522 Len=0 Win=8760
  0.00005      surfers -> sunset       TCP D=34018 S=5001     Ack=2341459085 Seq=1073031522 Len=0 Win=8760
  0.00043       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341459085 Len=1460 Win=8760
  0.00015      surfers -> sunset       TCP D=34018 S=5001     Ack=2341464357 Seq=1073031522 Len=0 Win=8760
  0.00042       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341467277 Len=1460 Win=8760
  0.00029       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341468737 Len=1460 Win=8760
  0.00019       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341470197 Len=892 Win=8760
  0.00011      surfers -> sunset       TCP D=34018 S=5001     Ack=2341464357 Seq=1073031522 Len=0 Win=8760
  0.00006      surfers -> sunset       TCP D=34018 S=5001     Ack=2341464357 Seq=1073031522 Len=0 Win=8760
  0.00005      surfers -> sunset       TCP D=34018 S=5001     Ack=2341464357 Seq=1073031522 Len=0 Win=8760
  0.00042       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341464357 Len=1460 Win=8760
  0.00016      surfers -> sunset       TCP D=34018 S=5001     Ack=2341465817 Seq=1073031522 Len=0 Win=8760
  0.00041       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341472549 Len=1460 Win=8760
  0.00017      surfers -> sunset       TCP D=34018 S=5001     Ack=2341465817 Seq=1073031522 Len=0 Win=8760
  0.20605       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341465817 Len=1460 Win=8760
  0.00015      surfers -> sunset       TCP D=34018 S=5001     Ack=2341471089 Seq=1073031522 Len=0 Win=8760
  0.00044       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341471089 Len=1460 Win=8760
  0.00015      surfers -> sunset       TCP D=34018 S=5001     Ack=2341474009 Seq=1073031522 Len=0 Win=8760
  0.00041       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341474009 Len=1460 Win=8760
  0.00031       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341475469 Len=1460 Win=8760
  0.00029       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341476929 Len=1460 Win=8760
  0.00012      surfers -> sunset       TCP D=34018 S=5001     Ack=2341476929 Seq=1073031522 Len=0 Win=8760
  0.00029       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341478389 Len=892 Win=8760
  0.00026       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341479281 Len=1460 Win=8760
  0.00013      surfers -> sunset       TCP D=34018 S=5001     Ack=2341480741 Seq=1073031522 Len=0 Win=8760
  0.00044       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341480741 Len=1460 Win=8760
  0.00026       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341482201 Len=1460 Win=8760
  0.00029       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341483661 Len=1460 Win=8760
  0.00012      surfers -> sunset       TCP D=34018 S=5001     Ack=2341483661 Seq=1073031522 Len=0 Win=8760
  0.00029       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341486581 Len=892 Win=8760
  0.00028       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341487473 Len=1460 Win=8760
  0.00012      surfers -> sunset       TCP D=34018 S=5001     Ack=2341485121 Seq=1073031522 Len=0 Win=8760
  0.00006      surfers -> sunset       TCP D=34018 S=5001     Ack=2341485121 Seq=1073031522 Len=0 Win=8760
  0.00035       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341488933 Len=1460 Win=8760
  0.00013      surfers -> sunset       TCP D=34018 S=5001     Ack=2341485121 Seq=1073031522 Len=0 Win=8760
  0.18509       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341485121 Len=1460 Win=8760
  0.00016      surfers -> sunset       TCP D=34018 S=5001     Ack=2341490393 Seq=1073031522 Len=0 Win=8760
  0.00044       sunset -> surfers      TCP D=5001 S=34018     Ack=1073031522 Seq=2341490393 Len=1460 Win=8760

 
 
 

Terrible 100Mbit Ethernet performance with 3Com Fast Etherlink XL

Post by John P. Eisenmeng » Thu, 24 Jul 1997 04:00:00


: Hi
: Hi have a Dell OptiPlex GXPro 200MHz running Solaris x86 2.5 update 10
: we have added a 100Mbit Ethernet card (3Com Fast Etherlink XL) and
: was hoping for some performance but it turned out that it was 100 slower
: than 10Mbit ethernet on receiving on transmitting it was acceptable
: performance.

We have found that very few of the fast ethernet card drivers have full
duplex support (even when documented in the driver's config options).  It
may be that the switch is operating in full duplex mode, but the PC card
isn't.  Looking at "netstat -i" do you see collisions or errors being
recorded on that interface?  If so, then try plugging into a half-duplex
port.

-John

 
 
 

Terrible 100Mbit Ethernet performance with 3Com Fast Etherlink XL

Post by Casper H.S. Dik - Network Security Engine » Fri, 25 Jul 1997 04:00:00



>When running ttcp between a Sun Ultra-170E and the PC we get
>Sun 170E    -> PC 100Mbit    94KB/s ( 0.752Mbit/s)
>PC 100Mbit  -> Sun 170E     3276KB/s (26Mbit/s)
>Sun 170E    -> PC  10Mbit*  1092KB/s ( 8.7Mbit/s)
>Sun 170E    -> Sun SS20    2730KB/s (21Mbit/s)
>*another GXPro with default builtin Ethernet card 10Mbit.
>snoop shows that there seems to be a 0.2s delay ever so often (surfers is the PC and
>sunset is the Sun)

That looks like a typical case of dropped packets and a retransmit
resulting.  (200ms is the default tcp_rexmit_interval_min).
(Changing the tcp_rexmit_interval_min value on teh Sunshould help throughput,
but the lost packets should be investigated)

Quote:>Any suggestions? is it a bug in the Etherlink III driver or is the card not upp to
>standard or what?
>An observation is that when you move the cable from a 10/100Mbit slot
>in a switch to a 10 or 100Mbit fixed slot the whole machine just hangs I guess that
>has to do with full duplex i.e. moving it from a switch that handles full duplex to
>one that does not.

What type of switches are you using?

They seem to be losing a lot of packets.

Casper
--
Expressed in this posting are my opinions.  They are in no way related
to opinions held by my employer, Sun Microsystems.
Statements on Sun products included here are not gospel and may
be fiction rather than truth.

 
 
 

Terrible 100Mbit Ethernet performance with 3Com Fast Etherlink XL

Post by WarpTwai » Sat, 26 Jul 1997 04:00:00


This is a known bug that has nothing to do with hardware.  It is a problem
with TCP standards, and there is a fix for it.  I assume the PC is running
a Microsoft OS.  Their implementation of TCP deviates from the standard and
is the cause of this slowdown.  Look for a fix on Sun's website soon.


Michael Zumpano

 
 
 

Terrible 100Mbit Ethernet performance with 3Com Fast Etherlink XL

Post by Schwitter Rued » Sat, 26 Jul 1997 04:00:00




> >When running ttcp between a Sun Ultra-170E and the PC we get
> >Sun 170E    -> PC 100Mbit    94KB/s ( 0.752Mbit/s)
> >PC 100Mbit  -> Sun 170E           3276KB/s (26Mbit/s)
> >Sun 170E    -> PC  10Mbit*  1092KB/s ( 8.7Mbit/s)
> >Sun 170E    -> Sun SS20    2730KB/s (21Mbit/s)

Have You set the SunCard on your Sun 170E to full-duplex/half-duplex ?
Is it an auto-detect Card (switch 10-100MB/s)?

See the manuals about the ndd command.
--
Regards Ruedi

----------------------------------------------------------------

Haselstrasse                            Fax: ++41 56 205 21 21
CH-5400 Baden
---------------------------------------------------------------

 
 
 

Terrible 100Mbit Ethernet performance with 3Com Fast Etherlink XL

Post by Rick Jon » Sat, 26 Jul 1997 04:00:00


: This is a known bug that has nothing to do with hardware.  It is a problem
: with TCP standards, and there is a fix for it.  I assume the PC is running
: a Microsoft OS.  Their implementation of TCP deviates from the standard and
: is the cause of this slowdown.  Look for a fix on Sun's website soon.

Precisely how does it deviate from the standard? And is that a de jure
or de facto standard? There has been a lot of vague discussion of this
topic with no concrete data at all. If there was a deviation in NT
from the de facto standard TCP behaviour why haven't we heard about
similar problems between NT and FreeBSD, or Digital Unix, or HP-UX, or
SGI Irix, or Linux?

rick jones

 
 
 

Terrible 100Mbit Ethernet performance with 3Com Fast Etherlink XL

Post by Russell Hamme » Tue, 29 Jul 1997 04:00:00




> : This is a known bug that has nothing to do with hardware.  It is a problem
> : with TCP standards, and there is a fix for it.  I assume the PC is running
> : a Microsoft OS.  Their implementation of TCP deviates from the standard and
> : is the cause of this slowdown.  Look for a fix on Sun's website soon.

> Precisely how does it deviate from the standard? And is that a de jure
> or de facto standard? There has been a lot of vague discussion of this
> topic with no concrete data at all. If there was a deviation in NT
> from the de facto standard TCP behaviour why haven't we heard about
> similar problems between NT and FreeBSD, or Digital Unix, or HP-UX, or
> SGI Irix, or Linux?

> rick jones


talking about is responsible for the percieved network
performance in the original thread, I would
like to comment on that issue since Rick Jones asked.

I believe the issue revolves around RFC2001 specifically how
the "slow start" algorithm is interpreted by the different
TCP implementations.  Who's right or wrong, I have no idea,
but it sounds like Sun is going to provide a fix/workaround
(depending on your point of view).

        see: http://ds.internic.net/rfc/rfc2001.txt

I may be wrong here, but FWIW I thought I would comment anyway.

Regards,
Russ

 
 
 

Terrible 100Mbit Ethernet performance with 3Com Fast Etherlink XL

Post by Rick Jon » Wed, 30 Jul 1997 04:00:00



: > : assume the PC is running a Microsoft OS.  Their implementation
: > : of TCP deviates from the standard and is the cause of this
: > : slowdown.  Look for a fix on Sun's website soon.
: > topic with no concrete data at all. If there was a deviation in NT
: > from the de facto standard TCP behaviour why haven't we heard
: > about similar problems between NT and FreeBSD, or Digital Unix, or


: about is responsible for the percieved network performance in the

Agreed - or it would have affected the 10 Mbit performance as well,
which is what twigged me in the first place :)

: original thread, I would like to comment on that issue since Rick
: Jones asked.

: I believe the issue revolves around RFC2001 specifically how the
: "slow start" algorithm is interpreted by the different TCP
: implementations.  Who's right or wrong, I have no idea, but it
: sounds like Sun is going to provide a fix/workaround (depending on
: your point of view).

Thank you - that is the first comment I've seen in a public forum that
was neither totally vague, or completely bizzarro (stuff about DF
bits). Why the press accounts and other stuff did not seem to mention
this puzzles me. (Or maybe I just missed them?)

It is ironic that as recently as a couple years ago, one of the most
common justifications I heard for changes in things networking was
"because that's the way Sun does it." (SunOS that is - stuff like
all-zeros broadcasting for one example.)

Times do seem to change.

rick jones

been on the receiving end of a "but the standard says foo not bar"
failure before...

these opinions are mine, all mine; HP might not want them anyway... :)

 
 
 

Terrible 100Mbit Ethernet performance with 3Com Fast Etherlink XL

Post by Tomas Stephans » Thu, 31 Jul 1997 04:00:00




>>When running ttcp between a Sun Ultra-170E and the PC we get
>>Sun 170E    -> PC 100Mbit    94KB/s ( 0.752Mbit/s)
>>PC 100Mbit  -> Sun 170E         3276KB/s (26Mbit/s)
>>Sun 170E    -> PC  10Mbit*  1092KB/s ( 8.7Mbit/s)
>>Sun 170E    -> Sun SS20    2730KB/s (21Mbit/s)
>>*another GXPro with default builtin Ethernet card 10Mbit.
>>snoop shows that there seems to be a 0.2s delay ever so often (surfers is the PC and
>>sunset is the Sun)
>That looks like a typical case of dropped packets and a retransmit
>resulting.  (200ms is the default tcp_rexmit_interval_min).
>(Changing the tcp_rexmit_interval_min value on teh Sunshould help throughput,
>but the lost packets should be investigated)
>>Any suggestions? is it a bug in the Etherlink III driver or is the card not upp to
>>standard or what?
>>An observation is that when you move the cable from a 10/100Mbit slot
>>in a switch to a 10 or 100Mbit fixed slot the whole machine just hangs I guess that
>>has to do with full duplex i.e. moving it from a switch that handles full duplex to
>>one that does not.
>What type of switches are you using?
>They seem to be losing a lot of packets.

Well I tried one switch and one hub. The switch is a BayStack 350T and
the 100Mbit hub is a GrandJunction 2800 with 8 100Mbit ports +
25x10Mbit To me it looks like the card and not the switch. As it works
between two Sun Sparcs ok.
All the machines are running Solaris 2.5

/Tomas
--
Tomas Stephanson
Ericsson Medialab
126 25 Stockholm
Sweden

 
 
 

1. 3com PCMCIA Card (3com Fast Etherlink XL 3C575-TX)

Hello,

        I'm a newbie with FreeBSD.

        I try to use an ethernet PCMCIA card (3com Fast Etherlink XL 3C575-TX).

        I succeeded in compiling the kernel for use of PCMCIA (another modem
card works fine).
        But the 3Com card doesn't seem to work.

        Is my /etc/pccard.conf obsolete? Is ther anything else to do?

Thanks...
--
Laurent LAVAUD
Universit Victor Segalen Bordeaux 2
Dpartement d'Informatique
146 rue Lo-Saignat           Fax : 05 56 99 13 60
33076 Bordeaux cedex          Tl : 05 57 57 15 03

2. FAT32 and TX chipset support

3. 3Com 3C905-TX-IBM Fast EtherLink XL Configuration in AIX 4.2

4. kernel syslog messages that I don't understand

5. 3COM Fast Etherlink XL 10/100Base-TX (3c905B-TX)

6. Splitting up /var/mail ??

7. 3COM 3C905-TX Fast EtherLink XL PCI Card

8. Linux and MPC860

9. 3Com Fast Etherlink XL Driver Needed for 2.0 kernel

10. ether: 3Com Fast EtherLink XL PCI-TX 10/100-BT ?

11. Correct driver code fro 3com Fast EtherLink XL?

12. 3Com Fast Etherlink XL cards on 7043-140