Plausible: zero collisions, zero errors?
Mark Landin
Recently I installed a 10/100Base-T Lan card in my HP9000 H50 running
HP-UX 10.20. The other end of the cable is plugged into a dedicated
100Mbit port on a switch. The connection is operating in full-duplex
mode.
I did a netstat -i today, and even though I have processed
approximately 25 million inbound and outbound packets, there have been
zero collisions and zero errors.
Is this plausible? Or is something hosed in the statistics collection
somewhere? (If only the LAN card had a collision indicator on it...)
----
Mark Landin
UNIX Sys. Admin, T. D. Williamson, Inc. (Standard disclaimer applies)
m5...@ix.netcom.com
"Before anyone passes judgment ... remember, we *are* in
the Arctic" .. Fox Mulder (prior to a physical exam)
M.C. van den Bovenkamp
Mark Landin wrote:
> I did a netstat -i today, and even though I have processed
> approximately 25 million inbound and outbound packets, there have been
> zero collisions and zero errors.
>
> Is this plausible? Or is something hosed in the statistics collection
> somewhere? (If only the LAN card had a collision indicator on it...)
When you're running a dedicated full duplex switch port, zero collisions
is not only plausible, but the expected result. You won't see any
collisions. Ever. Unless something's broken. Think about it: 2 devices
on a full duplex link can both receive and transmit at the same time (to
each other), and there aren't any others. Whether the zero errors is
correct I can't say, but it's not impossible.
Regards,
--
Marco van den Bovenkamp.
CIO EMEA Network Design Engineer,
Lucent Technologies Nederland.
Room: HVS BZK 38
Tel.: (+31-35-687)2724
Mail: bove...@lucent.com
Rich Seifert
In article <34F6CAA9...@lucent.com>, "M.C. van den Bovenkamp"
<bove...@lucent.com> wrote:
> Mark Landin wrote:
>
> > I did a netstat -i today, and even though I have processed
> > approximately 25 million inbound and outbound packets, there have been
> > zero collisions and zero errors.
> >
> > Is this plausible? Or is something hosed in the statistics collection
> > somewhere? (If only the LAN card had a collision indicator on it...)
>
> When you're running a dedicated full duplex switch port, zero collisions
> is not only plausible, but the expected result. You won't see any
> collisions. Ever. Unless something's broken. Think about it: 2 devices
> on a full duplex link can both receive and transmit at the same time (to
> each other), and there aren't any others. Whether the zero errors is
> correct I can't say, but it's not impossible.
>
As a first order approximation, a bit-error-rate of 1x10^-12 (reasonable
for a benign office-type environment) applied in Gaussian fashion to a
stream of frames with average length of 500 bytes (a typical result from a
mix of short and long frames) will yield a frame-loss rate of 40x10^-10, or
one error per 250 million frames. Your statistics look just fine.
--
Rich Seifert Networks and Communications Consulting
sei...@netcom.com 21885 Bear Creek Way
(408) 395-5700 Los Gatos, CA 95030
(408) 395-1966 FAX
"... specialists in Local Area Networks and Data Communications systems"
Look for: "Gigabit Ethernet: Technology and Applications for High-Speed
LANs" at: www.amazon.com/exec/obidos/ISBN=0201185539/6750-8607450-932225
Heribert Dahms
In <34F6CAA9...@lucent.com> bove...@lucent.com
(Marco van den Bovenkamp) writes:
: When you're running a dedicated full duplex switch port, zero collisions
: is not only plausible, but the expected result. You won't see any
: collisions. Ever. Unless something's broken. Think about it: 2 devices
: on a full duplex link can both receive and transmit at the same time (to
: each other), and there aren't any others. Whether the zero errors is
: correct I can't say, but it's not impossible.
Now change the example to 2 devices communicating each at sustained full
speed with a third (uplink) port. Any size of FIFO memory will fill up!
Then the switch either throws packets away or pretends collisions.
Bye, Heribert (da...@ifk20.mach.uni-karlsruhe.de)
Rich Seifert
In article <6d7gvm$5g2$2...@nz12.rz.uni-karlsruhe.de>,
DA...@ifk20.mach.uni-karlsruhe.de (Heribert Dahms) wrote:
If it is a full duplex link, it cannot "pretend collisions", since a full
duplex link cannot assert collision detect to the MAC. This is precisely
the reason we designed the flow control protocol in 802.3x; it allows
devices to throttle the station at the other end of a full duplex link in
the face of the type of congestion you describe. It prevents dropping
frames under transient overload conditions.
Of course, if the overload is sustained, then there is little you can do
about it. You have a configuration (capacity) problem in this case. This is
the reason for the deployment of high-speed uplinks on many switches.
Heribert Dahms
In <seifert-ya0230600...@192.0.2.3> sei...@netcom.com
(Rich Seifert) writes:
: In article <6d7gvm$5g2$2...@nz12.rz.uni-karlsruhe.de>,
: DA...@ifk20.mach.uni-karlsruhe.de (Heribert Dahms) wrote:
: > Now change the example to 2 devices communicating each at sustained full
: > speed with a third (uplink) port. Any size of FIFO memory will fill up!
: > Then the switch either throws packets away or pretends collisions.
: >
:
: If it is a full duplex link, it cannot "pretend collisions", since a full
: duplex link cannot assert collision detect to the MAC. This is precisely
: the reason we designed the flow control protocol in 802.3x; it allows
: devices to throttle the station at the other end of a full duplex link in
: the face of the type of congestion you describe. It prevents dropping
: frames under transient overload conditions.
I admit that I have not much knowledge of full duplex TP so far.
Maybe the switches signalling collisions do that only in half duplex mode.
Thinking again, TP uses only 2 pairs of the usually wired 4, while
an AUI cable uses 3 pairs (although I've never heard of full duplex AUI).
Don't know about the wiring of the faster MII (not yet gotten one).
Or I've confused even with hardware handshake on full duplex RS-232...
Bye, Heribert (da...@ifk20.mach.uni-karlsruhe.de)