From: Rik Guyler (rik@guyler.net)
Date: Thu Feb 05 2004 - 22:36:48 GMT-3
To be compliant with the IEEE specification you would need to flip all 4
pairs as GE uses 4 full-duplex pairs capable of 250Mbps in each
direction. However, a "normal" crossover (1-2,3-6) will function fine
with GE but you are reducing your full duplex pairs to only 2 so
throughput will be reduced by some factor. There is also a built-in
provision within the GE spec for MDIX negotiation but this is optional
so not all vendors support it. I haven't seen many Cisco devices
support this although I rather like the feature.
Rik
-----Original Message-----
From: Mike Williams [mailto:ccie2be@swbell.net]
Sent: Wednesday, February 04, 2004 9:03 PM
To: CCIELab@Groupstudy.com
Cc: goldchain@stlo.smhs.com
Subject: OT: 1000Base-T crossover working when it shouldn't?!?!?
Okay............. bear with me here, but this is strange, and Cisco
doesn't even have an answer for this........
The other day, one of our server people was setting up a server and
asked me to make sure his port was patched and active (we have patch
panels in the bottom of each server rack that runs back to the patch
panels by the switches). I checked and the port was already patched
over to our of our Cat4507s (with dual SupIVs with 48-port 10/100/1000
RJ-45 blades). I went and set the port for auto-speed/duplex (since the
server folks can't force their Gig NICs to 1000Mbps, we have to use auto
on both ends to get a 1000Mbps connection). He called about an hour
later asking me to check the port saying he didn't have a connection.
Sure enough it was up/down. He tried another patch cable (we're using
Cat6 patches cables with Cat6 runs). Again, nothing........ So he
tried a 3rd cable (which he didn't recognize as a crossover, but it was
a different color than our straight thru cables). Lo and behold....
BAM! The connection came up......... communications work fine, no
errors, etc......... Mind you, this run was previously used by another
server, and patched into our switch with a straight thru cable, so I
know the cable in the floor and the patch cable are good and straight as
they're supposed to be.
Then I started thinking...... 1000Base-T uses all 8 wires, not 4 like
10/100Mbps ethernet........ Having never made a 1000Base-T crossover by
hand, I started checking around, and finally found docs (even one on
Cisco's website) showing that for a 1000Base-T crossover, not only do
you cross 1/2 with 3/6, but you also have to cross 4/5 with 7/8. Of
course all of the Cat6 crossovers we ordered are "standard" crossovers
with only 1/2 and 3/6 crossed........ Then that made me realize
something major. All 4 of our 4507s connect to our 2 core 6509s with
these Cat6 crossovers!!! All of the ports (on both the 4507s and 6500s)
show a 1000Mbps connection full duplex.... no errors, or anything. And
according to our monitoring (Compuware) we have been pushing up to
600+Mbps across these gig links, so they seem they're working normally.
HOW IN THE WORLD could this be working?
So we had a theory that somehow these switchports were "autodetecting"
the crossed nature of the cables........ I checked Cisco's website, and
according to them, starting in IOS 12.1(13)E (native IOS on the 6500s)
there is a "mdix auto" command that only works on 3 particular blades,
none of which we're actually using. We do indeed have 12.1(13)E9, and
the command isn't there! (the Cisco document shows it being used in
enable mode, but I tried enable mode, global config mode, and even
interface config mode). Also, there is NO indication that the 4500s
support such a thing and given that that 6500s IOS only supports this
feature with 3 line cards, I doubt it was a high enough priority to put
in the 4500s IOS.
Which THEN brought me to this train of thought: EVEN IF...... EVEN IF
there were some magical way that the ports on both ends of these
connections (4500->6500s) detected that there was a crossed cable, how
could this succeed and work with the "standard" 10/100 Ethernet
crossover?!?! I mean, we're not even using true 1000Base-T crossover
cables, just ones with 1/2 and 3/6 crossed. We even began to think that
this magic ability was down to the pairs themselves, so that it could
actually use the 1/2 and 3/6 pairs but not the others, giving
500Mbps......... t'was an idea, but as I mentioned, we verified with our
monitoring software that (at night during backups in the datacenter) the
gig links are actually pushing 600+Mbps, so that blows the 500Mbps
theory..........
This is driving me crazy!!!!! Am I missing something here?......... I
can't complain too much, as all of the links seem to be
working.......... but this WILL drive me crazy if I can't find the
"logical" explanation for this..........
Any comments or information on this would be appreciated.
Mike W.
This archive was generated by hypermail 2.1.4 : Fri Mar 05 2004 - 07:13:47 GMT-3