dslreports logo
 
    All Forums Hot Topics Gallery
spc
Search similar:


uniqs
506
tired_runner
Premium Member
join:2000-08-25
CT
·Frontier FiberOp..

tired_runner

Premium Member

Curious gotcha between two 3750s

At work there are many questionable situations or setups that have been in place since before my time.

One of those I was tasked with cleaning up somewhat today.

Sparing all major details, the "clean-up" involved changing the link of one 3750 piggybacked to another one at a different location within the same campus, from an access port to a trunk. The link between the two is a CAT5 drop sufficiently long to prevent successful 100 full connection, it will only do 10 half.

But the trunk needed to happen anyway as there are additional devices about to be put on, and the subnet to which this was piggybacked is full.

Yes, I know. Please don't laugh.

With this background in mind, at first I tried using a 3750V2 with 12.55SE9 I believe. The other end of the piggyback is a 3750 running 12.2 SEB4. I staged this at my desk first using the same 3750 with same IOS and it trunked fine there. When I tried trunking it at location, the only thing the 3750V2 would do is blink amber and green, stating that it was trunking when it really wasn't. There was no connectivity.

I went back to my desk, grabbed an older 3750 with same IOS as the other one and bam, it's trunking.

Keeping in mind traffic choke, I limited the trunk to carrying three vlans only; two user traffic segments plus management traffic.

So.... How did it trunk with older gear and IOS on both ends but not mixing old and new? In my mind, I thought newer IOS and hardware would better tolerate weaker links while adhering to IEEE standards.

That's my head scratcher for the day......
cramer
Premium Member
join:2007-04-10
Raleigh, NC

cramer

Premium Member

Something tells me the cable length isn't the problem.
markysharkey
Premium Member
join:2012-12-20
united kingd

markysharkey to tired_runner

Premium Member

to tired_runner
I always thought 10Mb half duplex would fail at more than 100m because of the retransmit / hold down timers. 100Mb full fuplex would go further than 100m (subject to cable quality, EM etc etc) but how much further is not predictable.

I am going to assume static trunk, static speed and duplex and VTP mode transparent on both ends.
Maybe it's a protocol mis-match? Is one end doing ISL and the other DOT1Q?
Did you match the native VLAN?
tired_runner
Premium Member
join:2000-08-25
CT

1 edit

tired_runner

Premium Member

Static dot1q with nonegotiate, vtp transparent on both ends. Static 10 half on the switch it piggybacks from, auto on the other end. It wouldn't connect otherwise. Yes, native vlan is the same on both ends.
tired_runner

tired_runner to cramer

Premium Member

to cramer
What do you think it could have been?

I'd like to go back and set to 100 full.
markysharkey
Premium Member
join:2012-12-20
united kingd

markysharkey

Premium Member

Auto-MDX?
tired_runner
Premium Member
join:2000-08-25
CT

tired_runner

Premium Member

Auto MDX on one end only

Edit: oops.. Auto mdx on both ends
HELLFIRE
MVM
join:2009-11-25

HELLFIRE to tired_runner

MVM

to tired_runner
...able to supply the full config from both switches? Just to be sure... and save the A$$-U-ME-ing?

My thought would be checking device logs and "sh interface x/y" after you plugged it in -- as the old adage goes, PHYSICAL LAYER FIRST!

My 00000010bits

Regards

KA0OUV
Premium Member
join:2010-02-17
Jefferson City, MO

KA0OUV to tired_runner

Premium Member

to tired_runner
Du
said by tired_runner:

Static 10 half on the switch it piggybacks from, auto on the other end.

Classic duplex mismatch setup. Force both ends to 100Mb Full Duplex in the config. Here is my standard Cisco-to-Cisco trunk configuration:
interface GigabitEthernet1/9
description TRUNK to DEVICE_A [ge0/1]
switchport trunk encapsulation dot1q
switchport trunk allowed vlan 56,59
switchport mode trunk
switchport nonegotiate
speed 100
duplex full
spanning-tree portfast trunk

Some switches would not trunk on 10Mb Ports. Or worse, not trunk correctly. Force it on both ends, explicitly define your vlans, remember that some of the commands are default and hidden, and it should rock and roll.

TomS_
Git-r-done
MVM
join:2002-07-19
London, UK

TomS_ to tired_runner

MVM

to tired_runner
Urgh.

There was a time when I would have agreed, and manuall set everything to 100/full regardless of whether autoneg worked fine or not, but these days I am inclined to avoid hard coding speed or duplex unless absolutely necessary.

Autoneg should be left to do its own thing and achieve what it can achieve. If you try and force something to do something it cant you'll just end up with a crappy network.

Simple fact is that these days autoneg works fine. There was/may have been a time when that was not so true.
TomS_

TomS_ to KA0OUV

MVM

to KA0OUV
said by KA0OUV:

Classic duplex mismatch setup

Full duplex on one end and auto on the other, maybe. But it falls back to half duplex, so there shouldnt be any duplex mismatch in this kind of situation.

Also, why hard code a gigabit interface to 100mbit?
tired_runner
Premium Member
join:2000-08-25
CT
·Frontier FiberOp..

tired_runner to TomS_

Premium Member

to TomS_
I agree with you. I normally autoneg links between switches and it works without issue.

But these are decade-old 3750s I'm dealing with here, and although it autoneged fine at my desk, it's not liking it where it's needed to work.

Later today I'm going back there to try manual 100 full on both ends and see what happens. I expect it will balk and if such, I will leave it be.
tired_runner

tired_runner to TomS_

Premium Member

to TomS_
The only gig port on these switches is fiber. The switch it piggybacks from has fiber homerun to the core, but not the second one. Else this piggyback situation wouldn't exist.

I would like to get two clean Ethernet runs so I can at least do two-port Etherchannels, and later ask for fiber homerun to core. But when money is tight, this is what we do.
tired_runner

tired_runner to KA0OUV

Premium Member

to KA0OUV
said by KA0OUV:


Classic duplex mismatch setup.

So I went back and tried this on both ends:

no mdix auto
switchp nonego
switchp mod tru
switchp trun allo vla none
switchp trun allo vla add 100
switchp trun allo vla add 101
switchp trun allo vla add 102
switchp trun encap dot1q
spee 100
dup full
 

And still got interface down not-connected.

It only stays up with this on one end:
interface FastEthernet1/0/48
 switchp tru encap dot1q
 switchp tru allow vla 100,101,102
 switchp mod tru
 switchp nonego
 duplex half
 speed 10
 

And this on the other end:
interface FastEthernet1/0/1
 switchp tru encap dot1q
 switchp tru allow vla 100,101,102
 switchp mod tru
 switchp nonego
 



Suffice to say, this is enough for me to consider I'm working with either an extraordinarily long connection barely sufficient for IEEE 802.3 spec or there's EMI interference somewhere preventing this from hitting 100 full.

But I'm open for any ideas.
tired_runner

tired_runner to HELLFIRE

Premium Member

to HELLFIRE
said by HELLFIRE:

...able to supply the full config from both switches? Just to be sure... and save the A$$-U-ME-ing?

My thought would be checking device logs and "sh interface x/y" after you plugged it in -- as the old adage goes, PHYSICAL LAYER FIRST!

My 00000010bits

Regards

There's a lot for me to sanitize off the config. Anything in particular you want to see?

DarkLogix
Texan and Proud
Premium Member
join:2008-10-23
Baytown, TX

1 edit

DarkLogix to markysharkey

Premium Member

to markysharkey
said by markysharkey:

I always thought 10Mb half duplex would fail at more than 100m because of the retransmit / hold down timers. 100Mb full fuplex would go further than 100m (subject to cable quality, EM etc etc) but how much further is not predictable.

Well 10mbit uses 5v where as 100/1000 both use 3.3v so 10mb can go much further.

I'd see if you can find a tester that can show cable length then see how long that run is, then if its over 325ft make a cable that matches and coil it on your desk to test.

also maybe get some dumb switches and just see if they can make a connection.

if it is over 325ft then I'd see if its possible to replace with something like cat7, though doing that it might be possible to just do the fiber as the labor might not be to much more.
cramer
Premium Member
join:2007-04-10
Raleigh, NC
Westell 6100
Cisco PIX 501

cramer

Premium Member

(peak differential voltage)
10BASE-T: 2.2-2.8V
100BASE-TX: same as 10BASE-T (as best I can tell, I'm assuming 802.3 references to "11.1", which is 10BASE36, are a typo.)
1000BASE-T: "(0.75 V ± 0.83 dB)"

(The AC component is like 48v, but no one ever mentions that.)
HELLFIRE
MVM
join:2009-11-25

HELLFIRE to tired_runner

MVM

to tired_runner
Whole config... minus passwords and non-RFC1918 address... if possible tired_runner See Profile

I'm thinking faulty switchport itself -- so move it to a different interface if possible -- or as DarkLogix See Profile
said, test and redo the whole run.

My 00000010bits

Regards
chandom
join:2001-05-23
Tallahassee, FL

chandom to tired_runner

Member

to tired_runner
said by tired_runner:

Suffice to say, this is enough for me to consider I'm working with either an extraordinarily long connection barely sufficient for IEEE 802.3 spec or there's EMI interference somewhere preventing this from hitting 100 full.

But I'm open for any ideas.

Might be worth a shot with trying a cable-diag, plus it would give a reasonable estimate of cable length...

test cable-diagnostics tdr interface fa x/y/z
show cable-diagnostics tdr interface fa x/y/z

DarkLogix
Texan and Proud
Premium Member
join:2008-10-23
Baytown, TX

DarkLogix to cramer

Premium Member

to cramer
said by cramer:

(peak differential voltage)
10BASE-T: 2.2-2.8V
100BASE-TX: same as 10BASE-T (as best I can tell, I'm assuming 802.3 references to "11.1", which is 10BASE36, are a typo.)
1000BASE-T: "(0.75 V ± 0.83 dB)"

(The AC component is like 48v, but no one ever mentions that.)

Odd I recall reading in some ccna material around 2004 that 10base-t used 5v.
markysharkey
Premium Member
join:2012-12-20
united kingd

markysharkey

Premium Member

Never came across that in the CCNA.
Also voltage will have no effect on the hold down timers and they are the arbiter of whether a frame is re-transmitted or not, thereby causing a collision. I've never seen a switch have issues because of voltage not making it down a cable, and I've seen plenty of shonky cabling.

DarkLogix
Texan and Proud
Premium Member
join:2008-10-23
Baytown, TX

DarkLogix

Premium Member

Well remember voltage drops over distance, so higher voltage longer it'll go before its below the acceptable range.

Though I've seen 100mbit work fine at over 400ft (From an IDF to a warehouse desk) (we ran 4 cables in hopes it would work and replace the WET54G wifi bridge) connected the User's computer, printer, and sato printer, and a 4th for possible visitors.

Then again we made sure the cabling co used thicker gauge cable than normal (remember thicker the copper=less voltage loss over a given distance.)

Oh and one of the cables was used for the cisco phone then passing through to the computer.
tired_runner
Premium Member
join:2000-08-25
CT
·Frontier FiberOp..

tired_runner to markysharkey

Premium Member

to markysharkey
No voltage discussion in my CCNA books either, though it briefly discussed emanations and what they do with signaling on an Ethernet cable.

Back when I was dealing with Novell and NTLM's, I distinctly recall that whenever a connection had trouble on 100 full, falling back to 10 half got you online, however detrimental it would be to connection performance beyond seconds worth of bursts.

DarkLogix
Texan and Proud
Premium Member
join:2008-10-23
Baytown, TX

DarkLogix

Premium Member

said by tired_runner:

No voltage discussion in my CCNA books either,

It wasn't in the books it was in the class.

At lee college in Baytown they have a set of CCNA focused classes (they've changed a bit since I took them)

And it was in one of those classes that it was mentioned.
tired_runner
Premium Member
join:2000-08-25
CT

tired_runner

Premium Member

You know what else wasn't discussed in the books I read as far as I can remember? Fiber cable and connector types.

You'd figure that at least in CCNA R&S this should be taught but nope. Cisco whitepapers have been my friend.

DarkLogix
Texan and Proud
Premium Member
join:2008-10-23
Baytown, TX

DarkLogix

Premium Member

I bet the books also didn't demonstrate how tricky it can be to make a cable that can pass Cat5e cross talk specs.

that was one of the labs in the 1st class, make a cable then put it on a fluke cable quality analyzer rinse repeat till it passes. (just a little to much untwist and it would fail.)
HELLFIRE
MVM
join:2009-11-25

HELLFIRE

MVM

Don't remind me of that day... I think I ended up with quite a few blisters on my fingers trying to untangle, flatten, clip and crimp my cables till it worked...

...and instructor decided to make it "interesting" and have us make a straight, cross and rollover...

Fun Times [/sarcasm]

Regards

DarkLogix
Texan and Proud
Premium Member
join:2008-10-23
Baytown, TX

DarkLogix

Premium Member

Oh ya we had the same and had to make both ends.

if it failed XT or NEXT you'd have to figure out which end or both to redo.

/point at fluke cable quality tester
/shake fist with vigor

Oh and then even if it passed if the sheath wasn't in the connector enough for it to properly be held then it still failed.

So you'd have to cut the wires just the right length or start over again.
HELLFIRE
MVM
join:2009-11-25

HELLFIRE

MVM

's why they call it learning....

Regards