Untraceable collisions over UTP CAT5 - noise? 1

[nedstar1]

Hi.

I'm a net admin for a medium network, 92 stations. I've got 12-20 boxes each run to hubs, these hubs run to a backbone managed switch. In additon, I have another managed switch breaking up a long cable span.

Relevant info: One of my wintel boxes experiences scads of collisions when downloading, but not when uploading (or being downloaded from, I should say). TCP/IP of course, and everything is 100MB full duplex. I can log into my switch and monitor port counters for collisions, and through testing, I have arrived at the above conclusion.

I can't figure out what is causing my collisions. Cable span is about 50 feet. The machine operates fine on any other connection. I have not tried another machine on the line, but I have tested the cable, and the pinout tests fine. Pinging from a linux box returns no packet loss or significant delay. There is no traffic weighting enabled on the switch.

Here is what I suspect, and I am really looking for someone with EE experience who might be able to answer it. I suspect that the cable for this link may pass close to a flourescent ballast, and that this ballast is emitting enough EM noise to create false collisions. I cannot verify the cable route, as I inherited the network from someone else. Also, our helpful maintenance guys have installed insulation above the suspended ceiling, so I can't see anything up there anyways.

Is there a way to test for line noise with a PC? Is there a good tool for this type of troubleshooting? Is there some obvious solution that I can't seem to grasp? I'm stumped. I could use any and all advice.

TIA
nedstar1

 

[Guest_imported]

Hi Nedstar1

I'm no expert, but here's a few things to try...

When the office is empty, try turning off all lights in the vicinity of the likely UTP cable run. Does this make any difference?

Get your network cable maintenance guy to make you up a temporary 50ft (or whatever length) lead of UTP and punch the usual RJ-45 terminal boxes on either end. Again, do this when the office is empty (Health and Safety at work etc). Lay the cable temporarily across the floor from the relevant PC to the hub/switch, removing the fixed cable patches from the wall socket and the patch panel, and patch this new lead in place. Does this make any difference? This will go some way towards establishing if the installed cable is picking up an unwanted electrical field. I would also re-check existing cable punchdowns in the wall socket and patch panel, and try different patch leads.

Good luck!


ROGER - GØAOZ.

 

[MasterRacker]

The noise could be from the cable, poor patch cables, poor punches in the patch panel or wall jack, etc. As Roger says, you need to try and isolate it. I would go about it a little different though.

Swap out the patch cables on both ends of the link. Try it with the offending computer and another computer. Try both computers on a known good link. If both have a problem on the questionable link and both don't on the good one, you know it's some component of the link. Your time is important and costs your company money also though, so I wouldn't go any farther than this. Time is far more expensive than wiring components so at this point I would simply have the connection re-run with new components following a known good CAT5 compliant path.
Jeff

I haven't lost my mind - I know it's backed up on tape somewhere ....

 

[MWM]

Nedstar1
Step One: See
A person can know a lot about the likely source of problems in about ten minutes of looking at your physical plant. This is only step one, but it is step one.

This perspective on cable troubles is a focus on the quality of the installed plant. Yes, I do believe some cable brands and termination hardware are better than others. Lucent-Avaya-whoever they are now, is very good and Panduit is very good for hardware. Belden in my favorite in cable. And, yes, it costs more. So less expensive hardware, such as Leviton, is, one of the usual suspects around the scene of the crime.

Regarding noise propagated from ballasts and fixtures, tests by Belden at a BICSI conference a couple of years ago showed no measurable noise induced into their cable. You might get a copy of the monograph from Belden. Of course, unusual circumstances, such as a ballast going bad, might result in enormous amounts of radiated noise.

Quality again, in the pathway used for the cable. There is a spec for placement near such noise and power sources. BICSI and the Assoc. of Cabling Professionals can refer you to appropriate EIA/TIA installation guidelines. I mention these guidelines so you can, if necessary, defend yourself by pointing to suspect "inherited" cable plant and referring to specific guidelines. Use real numbers: bosses are impresesed with that.

All the above is indicative of problems, not definitive. Testing can yield definitive results.

Step Two: Test
Electrical faults? A simple reversed polarity receptacle is no longer acceptable. I use an Ideal Industries receptacle tester, about $300. It even reads out the impedance to ground. Clean power needs to be verified, and it is quickly done--if you have the tools.

Regarding test tools, you may have a hard time getting funding for a good cable test tool, but it is the only objective reading. They can certify cable to standards, which is where the rubber meets the road. These guys are $5,000 to $6,000. You cannot eyeball this stuff. This is a definitive test: it passes or it doesn't.

Quality again in the hubs you mentioned. Old? Behind some ghost-like problems is a piece of prehistoric hardware. Hubs and switches are much better now than even a couple of years ago. Pay more, sleep more. You cannot see these problems. You might stress the link or port using definable traffic loads and watch the major criteria--usage, collisions, errors, etc. There are guidelines for these things. Easy and definitive if you have a tester.

Nic cards? Is the one in the suspect machine a quality item? The managed ones can be surprisingly helpful in narrowing the scope of the problem. My Fluke tester gives the voltage off the NIC card, so a failing NIC with marginal (but usually functioning, sort of) levels can be spotted. A bad patch jack can also provide a high-resistance contact that still works, sort of.

One cannot eyeball this stuff. Pay more, sleep more.

Your company cannot expect you to troubleshoot this stuff with your bare hands.

Best
Mike, RCDD

 

[nedstar1]

This is all excellent information - let me try to address some of the questions raised above. All of our receptacle hardware is made by Belkin, and it seems to be quite high quality. In additon, I replaced the receptacle in this location to rule out defective receptacle. Our patch cables are nedstar1-made. The cable is inherited, so I cannot verify it's maker, but I THINK it is made by a company called Condumex, which may be an issue. I have found defective spans of this in the past. Our hubs and switches are all made by HP, all 24 port, all less than 3 years old, and all 10/00 autosensing. I haven't had any problems with them, except for one dead/bad port on one old switch that resided out in the open in our production facility. We work in plastics, and I imagine the dust from the processes worked its way in and wreaked some habit. This hub is now in a positive-pressure filtered remote equipment box.

I have not tried operating the machine with the lights off, but this is not a bad idea, G0AOZ. However, per MWM's post, it should not even be an issue. I relocated the machine to a known good drop (mine) and it tested fine. Running ethereal on a linux box, there were a few packet errors, but greatly diminished from its permanent location.

The NIC on the box in question is built-in to the motherboard (Dell) and is based on a 3com chipset.

I have swapped out and tested all of the patches in the drop, from backbone to desktop, and even would up dropping a few $$$ on some pre-made cables.

I am going to take all of the advice above into consideration. I am keen on routing our network cabling according to established guidelines, but the cost is a large issue. Due to a poor season for our business, we are cash-poor, and my budget has been severely curtailed for the remainder of the year. (last year, we retired 24 PCs. This year, only three) Ergo, no chance at all of a Fluke-quality tester. Zero chance.

I have the good fortune to work in an environment of mutual trust - if I tell them there is a problem, and that we HAVE to fix it, then we will fix it, no red tape at all. My manager trusts my judgment implicitly, and he reports directly to the CFO and CEO, who are both approachable and reasonable people.

We do things differnetly here than anywhere I've ever worked. All of the cable is run by our maintenance department, and it pretty much is linear from receptacle to network room. I do all of the terminations for my own peace of mind. I really need to get my hands on the EIA/TIA specs for cable routing - I am a completely self-taught, so official documentation is very helpful to me.

Thanks to everyone for all the great advice - if you have anything more to add, the forum will notify me by email, so just reply here. Thanks again.

Nedstar1

 

[MWM]

Nedstar1

The formal spec for cable installation--which is to be referenced in the upcoming 2003 NEC revision--is called TIA/EIA 568A-1 thru 3, and is only available through Global Engineering Documents and costs a few hundred dollars for the set. There are sections on grounding, documentation, building entrance, horizontal cabling, etc.

However, for much less BICSI has a Cabling Installation Manual (

http://www.bicsi.org/installc.htm)

and a Telecommunications Distribution Methods Manual. Both excellent for their purposes. These are the practical details of installation in support of the TIA 568 standards.

Best
MWM

 

[sleew]

Hi Nedstar,
All advices mentioned above are O.K.Try them one by one and dont miss any.Nic,cable,panels,hubs,switches-it's your route.All of this components besides switch you have in more than one pieces so troubleshooting should not be an problematic issue.Just one add-on:Try OEM support tools for your equipment.Find out brandmarks for each component, go to internet and try to download some sw tools for diag.Then,
search forums,white papers,customers advisores,know.base, go to OEM forums to find out problems with certain models.Digg out boxes from basement,find out some CDs and kits.Use standard and OEM methods.If you have Windows NT 4.0 or W2000S use Network monitor(it's a great tool) to noose packets,examine packets to
see is there nonsensical content.Maybe some extra information.Also try some protocol analyser from internet for same purpose.Search sensless data is it in headers or in data it self.Search for CRC and NDR notifications.Borrow
Lan analyser somewhere,search for transformers locations in your buildings(they made me a problem with monitors)
USE RESOURCES,ELIMINATE PROBLEMS, YOU MUST SUCCESSED!


Sleew
Compaq & Microsoft TS
sleew@infosky.net

 

[kcottrcdd]

If you suspect the Cat5 cable..... It should be tested with a Level 2 cable scanner. This will not only check for continuity and polarity - but checks all sorts of other things. If it passes the scanner test..... It should support clean 100MB traffic. And the problem will be elsewhere. These scanners, Fluke DSP4000, Wirescope, Microtest Omni, and others are expensive. But your local cabling contractor should have one. It may be worth a couple of bucks to have the cable tested to eliminate it.

Keith M. Wolcott, RCDD

 

[Warped]

First of all try setting your NIC to half duplex and try exactly the same thing. If it fails then try running at 10 mb and do the same. If it passes you could try taking the nic settings and force a 100mb connection, some switches under a load will break the 100mb connection and try and renegotiate at whatever it can. Hewlett Packard said that is a common issue with hubs and cheaper switches... But this is only the start of eliminating these things...it verywell could be a defective card....

Good luck.

Vaughn

 

[MWM]

Nedstar1

What's the latest on your collision problem?

If all the various avenues to solution advanced so far have yet to lead to a solution, then there is always the last effort: PUNT.

Unfortunately, life sometimes seems to hold back the resources to do the job.

One viewpoint-- change the machine, change ports, change anything to get it to work--then console yourself with the thought that someday you may learn what the problem actually was.

Best,
MWM

 

[nedstar1]

The issue has been resolved. I went up into the celing to try to see if there were an obvious problem with the cable, but there was not. The only things I saw were that the installers had coiled excess wire above the ceiling level, rather than cutting the cable to length. I suspected there could be some crosstalk from this. Ultimately, I did what I normally have to do. Rather than spend a great deal of time addressing the problem, I replaced the wire. While this gave me no satisfaction in determining the problem, the issue was instantly resolved, and that was my primary goal.

I appreciate the help you all offer here - I value these forums a great deal, mostly because of all the good advice.

Thanks everyone.

 

[cowboy405]

So Roger was right. Simple test. Run an alternate cable to the
offending node. If collisions disappear, cable is the culprit. Could
have caught it quick. Rely on tech support that suggests simple
solutions first, but MAKE 'EM STAY ON THE LINE WHILE YOU
TEST THEM, hehe.


Dennis