lots of collisions, is this normal?

lots of collisions, is this normal?

Post by Armin Krawinke » Mon, 28 Oct 2002 14:26:02



hi ng

i set up a fileserver using nfs (only linux clients) which is
connected to a hub. all clients are also connected to this hub, it is
some 3com officehub afair (10/100). the server and the clients all
have rtl and tulip chips and i had no problems configuring them
(everything runs at 100 MBit).

when i transfer large amounts of data i can see the hub's collision
led be lit almost all of the time...this isnt normal is it?

also check out the collisions field in ifconfgs output:

eth0      Link encap:Ethernet  HWaddr 00:50:FC:56:8D:06
          inet addr:192.168.0.3  Bcast:192.168.0.255  
Mask:255.255.255.0
          inet6 addr: fe80::250:fcff:fe56:8d06/10 Scope:Link
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:120900930 errors:33 dropped:0 overruns:0 frame:0
          TX packets:125298098 errors:6 dropped:0 overruns:10
carrier:12
          collisions:12206862 txqueuelen:100
          RX bytes:4137227031 (3.8 GiB)  TX bytes:2308886771 (2.1 GiB)
          Interrupt:11

the funny thing is that i get quite good transfer rates despite the
large amount of collisions. copying large files i get like 77 MBits/s
peaks and 50 MBits average (measured with nload).

where shall i start tracing the problem?

i am using debian sid (yes sid and this is not the prob since it was
the same using woody :)) nfs-kernel-server and nfs-common version
1.0.2-1. on the server i use kernel 2.4.18 with the sgi xfs patch and
the clients use 2.4.19 with the xfs patch.

any help would be appreciated

--
Rgds,
take a look at http://www.dtch.org
if you have a minute or two...

 
 
 

lots of collisions, is this normal?

Post by Michael Heimin » Mon, 28 Oct 2002 16:08:59



[..]

Quote:> carrier:12
>           collisions:12206862 txqueuelen:100
>           RX bytes:4137227031 (3.8 GiB)  TX bytes:2308886771 (2.1
>           GiB) Interrupt:11

> the funny thing is that i get quite good transfer rates despite
> the large amount of collisions. copying large files i get like 77
> MBits/s peaks and 50 MBits average (measured with nload).

Collisions are normal on a loaded hubed subnet, exchange it with a
switch and the problem should be solved, or just ignore the errors;)

Michael Heiming
--

figure it out

 
 
 

lots of collisions, is this normal?

Post by James Knot » Mon, 28 Oct 2002 23:41:31



> when i transfer large amounts of data i can see the hub's collision
> led be lit almost all of the time...this isnt normal is it?

> also check out the collisions field in ifconfgs output:

> eth0      Link encap:Ethernet  HWaddr 00:50:FC:56:8D:06
>           inet addr:192.168.0.3  Bcast:192.168.0.255
> Mask:255.255.255.0
>           inet6 addr: fe80::250:fcff:fe56:8d06/10 Scope:Link
>           UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
>           RX packets:120900930 errors:33 dropped:0 overruns:0 frame:0
>           TX packets:125298098 errors:6 dropped:0 overruns:10
> carrier:12
>           collisions:12206862 txqueuelen:100
>           RX bytes:4137227031 (3.8 GiB)  TX bytes:2308886771 (2.1 GiB)
>           Interrupt:11

> the funny thing is that i get quite good transfer rates despite the
> large amount of collisions. copying large files i get like 77 MBits/s
> peaks and 50 MBits average (measured with nload).

> where shall i start tracing the problem?

By understanding that collisions are a normal part of ethernet, when using
an unswitched hub or coaxial cable.  The more computers and more data
transfered, the greater the likelyhood of collisions.  Now according to the
figures you listed, you've got about a 10% collision rate.  Depending on
your network load, this may be normal.  If you are running a switched hub,
there may be a duplex conflict somewhere, as you'd not normally have
collisions with a switched hub.

--

Fundamentalism is fundamentally wrong.


james.knott.

 
 
 

lots of collisions, is this normal?

Post by Armin Krawinke » Tue, 29 Oct 2002 00:11:51


Well thanks, i just wasnt aware that such a big amount of collisions
is to be considered normal using a hub...

i will exchange the hub and use a switch to get the most out of it.

thank you very much

 
 
 

lots of collisions, is this normal?

Post by James Knot » Tue, 29 Oct 2002 08:18:30



> Well thanks, i just wasnt aware that such a big amount of collisions
> is to be considered normal using a hub...

> i will exchange the hub and use a switch to get the most out of it.

I recall seeing a graph several years ago, about throughput over ethernet.  
It appears to peak as loading aproaches about 35% or 40%.  Above that,
throughput actually decreases, due to all the collisions.

That problem disappears with switched hubs, though they have a similar
problem, referred to as "funneling", where one port gets overwhelmed with
traffic from several others.

--

Fundamentalism is fundamentally wrong.


james.knott.

 
 
 

lots of collisions, is this normal?

Post by Donald Beck » Tue, 29 Oct 2002 14:24:39





>> Well thanks, i just wasnt aware that such a big amount of collisions
>> is to be considered normal using a hub...

>I recall seeing a graph several years ago, about throughput over ethernet.  
>It appears to peak as loading aproaches about 35% or 40%.  Above that,
>throughput actually decreases, due to all the collisions.

Please don't repeat this myth.
This was a common mis-truth spread by token ring salepeople a decade
ago.
An Aloha-type network exhibits this behavior, but Ethernet does not.
It's very common to get 95% of the available bandwidth on a shared
Ethernet.

The commonly referenced paper on this almost a decade and half old:

Measured Capacity of an Ethernet: Myths and Reality (1988)
 David R. Boggs, Jeffrey C. Mogul, Christopher A. Kent
 Computer Communication Reviews

http://research.compaq.com/wrl/publications/abstracts/88.4.html

--

Scyld Computing Corporation             http://www.scyld.com
410 Severn Ave. Suite 210               Beowulf Clusters / Linux Installations
Annapolis MD 21403

 
 
 

lots of collisions, is this normal?

Post by Raqueeb Hass » Tue, 29 Oct 2002 22:07:03


well, I've been using this normal shared hub for half of my life ..
(reason -- why spending more when I get what I wanted) and saw pretty
less collisons ....

collision depend on ethernet cards too .... and bad drivers are making
it real bad (like some windows drivers on generic chipset).

raqueeb hassan
bangladesh

 
 
 

lots of collisions, is this normal?

Post by Raqueeb Hass » Tue, 29 Oct 2002 22:07:02


well, I've been using this normal shared hub for half of my life ..
(reason -- why spending more when I get what I wanted) and saw pretty
less collisons ....

collision depend on ethernet cards too .... and bad drivers are making
it real bad (like some windows drivers on generic chipset).

raqueeb hassan
bangladesh

 
 
 

lots of collisions, is this normal?

Post by James Knot » Wed, 30 Oct 2002 11:21:15



>>I recall seeing a graph several years ago, about throughput over ethernet.
>>It appears to peak as loading aproaches about 35% or 40%.  Above that,
>>throughput actually decreases, due to all the collisions.

> Please don't repeat this myth.
> This was a common mis-truth spread by token ring salepeople a decade
> ago.
> An Aloha-type network exhibits this behavior, but Ethernet does not.
> It's very common to get 95% of the available bandwidth on a shared
> Ethernet.

Actually, I heard about it first for Aloha and then by a DEC rep, talking
about DECNET (long before I knew about TR), which we were getting ready to
install for a bunch of VAX 11-780s.  Also, I doubt you'd get anywhere near
95% on a large unswitched network with several busy nodes.  I have also
read on many occasions that a 16 Mb token ring approaches the throughput of
a 100 Mb ethernet.  Here's a quote from "Scott Mueller's Upgrading and
Repairing Networks" 2nd ed., by Terry Ogletree, pg 174.  "In theory, you
might expect that a network operating at 10 million bits persecond would
actually be capable of transmitting that much data on a continuous basis.  
However, that is not hte case.  In most ethernet networks, the actual
utilization rate is only 40 percent before performance begins to degrade
rapidly.  As utilization rises, so do collisions."  I have often seen
similar comments in other networking books.

Also, since you mentioned token ring, it allows frames approximately 10x the
size of those on ethernet, so you also gain efficiency, due to reduced
overhead.  I used to have my computer set for a 4K MTU.  Token ring can
happily run indefinitely at 100% utilization.

--

Fundamentalism is fundamentally wrong.


james.knott.

 
 
 

lots of collisions, is this normal?

Post by Baho Uto » Thu, 31 Oct 2002 07:59:06




>> when i transfer large amounts of data i can see the hub's collision led
>> be lit almost all of the time...this isnt normal is it?

>> also check out the collisions field in ifconfgs output:

>> eth0      Link encap:Ethernet  HWaddr 00:50:FC:56:8D:06
>>           inet addr:192.168.0.3  Bcast:192.168.0.255
>> Mask:255.255.255.0
>>           inet6 addr: fe80::250:fcff:fe56:8d06/10 Scope:Link UP
>>           BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1 RX
>>           packets:120900930 errors:33 dropped:0 overruns:0 frame:0 TX
>>           packets:125298098 errors:6 dropped:0 overruns:10
>> carrier:12
>>           collisions:12206862 txqueuelen:100
>>           RX bytes:4137227031 (3.8 GiB)  TX bytes:2308886771 (2.1 GiB)
>>           Interrupt:11

>> the funny thing is that i get quite good transfer rates despite the
>> large amount of collisions. copying large files i get like 77 MBits/s
>> peaks and 50 MBits average (measured with nload).

>> where shall i start tracing the problem?

> By understanding that collisions are a normal part of ethernet, when
> using an unswitched hub or coaxial cable.  The more computers and more
> data transfered, the greater the likelyhood of collisions.  Now
> according to the figures you listed, you've got about a 10% collision
> rate.  Depending on your network load, this may be normal.  If you are
> running a switched hub, there may be a duplex conflict somewhere, as
> you'd not normally have collisions with a switched hub.

Then how is this explained?  eth0 is the internet side of my server which
has an up time of, 6:02pm  up 50 days, 21:48,  22 users.  


eth0      Link encap:Ethernet  HWaddr 00:60:08:A7:05:65
          inet addr:24.93.123.6  Bcast:255.255.255.255  Mask:255.255.252.0
          UP BROADCAST NOTRAILERS RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:20908917 errors:0 dropped:0 overruns:0 frame:0
          TX packets:10202357 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:100
          RX bytes:2046463702 (1951.6 Mb)  TX bytes:3936938014 (3754.5 Mb)
          Interrupt:10 Base address:0xde00

eth1      Link encap:Ethernet  HWaddr 00:60:08:39:61:32
          inet addr:192.168.1.1  Bcast:192.168.1.255  Mask:255.255.255.0
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:21994880 errors:0 dropped:0 overruns:0 frame:0
          TX packets:15260570 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:100
          RX bytes:3757985556 (3583.8 Mb)  TX bytes:1948698192 (1858.4 Mb)
          Interrupt:11 Base address:0xdc00

 
 
 

lots of collisions, is this normal?

Post by James Knot » Thu, 31 Oct 2002 09:27:35



> Then how is this explained?  eth0 is the internet side of my server which
> has an up time of, 6:02pm  up 50 days, 21:48,  22 users.

Do you have a switched hub?  How many computers on your net?  How busy is
it?

--

Fundamentalism is fundamentally wrong.


james.knott.

 
 
 

lots of collisions, is this normal?

Post by Baho Uto » Fri, 01 Nov 2002 08:35:25




>> Then how is this explained?  eth0 is the internet side of my server
>> which has an up time of, 6:02pm  up 50 days, 21:48,  22 users.

> Do you have a switched hub?  How many computers on your net?  How busy
> is it?

I have a switched hub on the eth1 side of the house, on the internet side
of the house who knows as I share than side (the eth0 side) with an
unknown number of people as it is a shared network.  On the eth1 or
internal side it can be from 1 to 25 depending on the circumstance.  As
you can see it did 2 to 4 gigabytes of traffic in 50 days. So it is
somewhat busy.
 
 
 

lots of collisions, is this normal?

Post by James Knot » Fri, 01 Nov 2002 09:23:06



>>> Then how is this explained?  eth0 is the internet side of my server
>>> which has an up time of, 6:02pm  up 50 days, 21:48,  22 users.

>> Do you have a switched hub?  How many computers on your net?  How busy
>> is it?

> I have a switched hub on the eth1 side of the house, on the internet side
> of the house who knows as I share than side (the eth0 side) with an
> unknown number of people as it is a shared network.  On the eth1 or
> internal side it can be from 1 to 25 depending on the circumstance.  As
> you can see it did 2 to 4 gigabytes of traffic in 50 days. So it is
> somewhat busy.

Well, as I mentioned in another note, you don't get collisions with a
switched hub, because you're not sharing anything where collisions can
occur.   You've even got separate send and receive paths, so that you can
have data flowing both ways at the same time.

--

Fundamentalism is fundamentally wrong.


james.knott.

 
 
 

1. collisions, frame and carrier errors..normal?

Just setup a dual NIC RedHat Linux firewall.  I noticed the internal
interface has some collisions and even worse, frame and carrier errors.
Here is the output today:
          RX packets:165483 errors:0 dropped:0 overruns:0 frame:6
          TX packets:95926 errors:0 dropped:0 overruns:0 carrier:5
          collisions:2181 txqueuelen:100
Should I be overly concerned about these errors?  I figure the
collisions are <1% of the total tx/rx most of the time but what about
the frame and carrier errors?  Also, I noticed that the # of collisions
seem to shoot up suddenly versus gradually.  Maybe a NIC spewing bad
packets?  Incidentally, we have a WAN without a switch installed, so I
expect some collisions but what is acceptable?
TIA,
Matt

2. Telnet and missing "file descriptors"

3. how many collisions are normal?

4. minicom help

5. What is a normal network collision rate??

6. Will a Alpha 4000 power supply fit an Alpha 1000

7. Collisions: What is 'normal'?

8. Solaris "format -e" SCSI expert mode?

9. lots of collisions [local NETWORK]

10. A lot of collision on my network !

11. Problems with Intel eepro100 : lots of carrier-errors and collisions

12. lots of hub collisions with FA-311 and DFE530-TX

13. lot's of collisions on 3c509 after 40k