Sorry, it's yet another "Matt's stumped by networking" question.

I'm doing work for my church, and they have a pretty make-shift network of switches supplying the network to their offices. Here's the general layout:

Server closet switch [#1] <-200ft-> switch #2 in maintenance closet <-60ft-> switch #3 in main office <-> office computers

Last Friday they started complaining about slow internet speeds, and they're right. It's ridiculously slow. At times it just seems like it's frozen, and then sometimes it'll finish loading but others it'll just stop.

So, I started eliminating possibilities (based on ease of testing ). First I tried a computer connected to switch #1. No problems at all, full 30Mbps connection from our provider.

I then tried plugging into switch #3. Slowness.

I then plugged into switch #2. Slowness.

I then tried using the line feeding switch #2 (the 200ft line). No problems.

I plugged the switch back in with only the 200ft line and my laptop connected. Slowness. I then tried a new switch with the same setup. Still slowness.

I'm going to try a brand new switch tomorrow, just in case, but I wanted to see if you good people might have some ideas as to what's wrong here. I'd really appreciate it. I really want to get these folks up and running again, as they were broken into and robbed last week and it isn't helping moral to have no internet access. Thanks for any help you can offer.

I'd guess ethernet autonegotiation problems. Are they managed switches? Can you force them to 100Mbps/full-duplex?

Try using switch #2 or #3 as switch #1. Only connect the computers currently connected to switch #1 is it slow?

Are the computers connected to switch #1 slow when switch #2 or #3 is connected? Did disconnect the computers connected to switch #2 and #3? It could be something running on multiple computers, bad network cards, a bad switch or bad cables.

Maybe a surge or lightning hit the network. I had a switch of mine get hit by lightning and it would work for a few days and then only send out bad packets to everything connected.

Perhaps you have a loopback between one or more switches and the network is saturated?

I inherited this network setup, and it's essentially a larger home network, so no, they're mostly using little Linksys switches.

I really can't swap switch #1 with anything because it's back in the server rack we just set up when they added on to the building. Essentially the hub of the network moved into this new construction, and an underground Cat6 cable (I think they put in Cat6 and not Cat5e) feeds a switch in a maintenance closet (switch #2). That switch in turn goes up to an office where the cable modem and a router used to sit before it was moved into the new construction. Because it wasn't approved to run new drops to the offices, I basically swapped out the modem and router for a switch.

This was all working just fine until a week ago!


Is there a way I'd diagnose that?


I did try disconnecting everything but switch #2 and plugged my laptop into it, and got slowness. I also tried the same but replaced switch #2 with a completely different switch, and I still got slowness. But that was an older switch, so I'm going to try a new one today.


By the way, I gather that the effective length of Cat5e is 100 meters, but isn't it boosted again if a switch is placed at the end of it? Do I get another 100 meters? Or is it lessened significantly? I'm not positive of the distance between #1 and #2, but I know it's not more than 100 meters.

Move the 200' cable to a *different* port on switch #1 and then do your (great!) tests again. Then try a different port at the other end (switch #2).

And triple check that the two plugs on that 200' cable are wired (1) identically, and (2) correctly. I had very similar problems here once, with a cable that didn't have the end plugs quite right.

-ml

Since everything was working fine and is now failing to work, there's a possibility that something went wrong with one of the long cable runs. Rats gnawing through cables, that sort of thing.

100m is the maximum distance between any two switches or between any switch and an end-device. Assuming perfect connectors and cables etc..

If you add a second switch, then it is possible to get another 100m from that switch. And so on. But only with *switches*, not hubs. And some brands/models give smaller values in their specifications, such as 100' instead of 100m between switches.

Cheers

Thanks so much for the help, guys. Here's some more diagnostics:

- I tried switching the 200' run to a different port on switch #1, and nothing has changed

- the cable ends LOOK good to me, but what do I know?

- I've gotten some rough speed test results from that test that Speakeasy has, and here's my results:
-- when plugged into switch #1, I get 30/5Mbps, which jives with what we should be getting from Cox, our ISP
-- when plugged directly into the end of the 200' run, I get around 18/5, which I figured was reasonable given the distance, is that correct? Or should it still be 30/5 at that point?
-- when the switch is added and only my computer plugged in, I get varying results between 300KBps and 1.6Mbps down, but the whole while I still get 5Mbps up.

Any more ideas?

*edit*
Oh, and putting in a brand new switch in place of switch #2 does nothing at all.

By the way, I'm not using auto-sensing switches, so these Linksys models have an uplink port. I want to make sure I have things set up properly. Am I right in assuming that the switch that is farther away from the router is the one that gets connected to its uplink port? In other words, in my above example lets say that switches #2 and #3 are 8-port switches. I would run a cable from port 1-7 on switch #2 to the uplink port on switch #3, correct? And the 200' run is plugged into the uplink port on switch #2, right?

I was under the impression that this was how it was done. Just to make sure (because I'll try anything at this point), I've reversed it and even not used the uplink port. This either didn't help or just shut down the network entirely, as you might assume

Just so -- on a switch (as opposed to router), the only difference between an "uplink" port and a normal port is that the socket is wired reversed, so you don't need a cross-over cable to patch it to a non-uplink port on another switch. When connecting two switches together, you connect a normal port on one to an uplink port on the other; it doesn't matter in terms of functionality which way round you do it, but it'd be less confusing for the next network administrator if the uplink port is the one that leads closer to the actual Internet.

And just out of interest, what on earth (or otherwise) does a church need a 30/5Mbps Internet connection for? That's a lot more than Sigmatel Cambridge ever had...

Peter

Move switch 2 next to switch 1. Plug it in and try again.

Excellent idea, that's one diagnostic I hadn't tried. Switch 2 next to switch 1 worked fine.

I even re-capped the far end of the 200' run in case that was the problem. No change.

These test would seem to indicate that the 200' run is the problem. Well, technically the sum of these tests appear to indicate that nothing is a problem. When I'm plugged right into the end of that 200' run, everything is fine. When I attach a switch, everything slows down to a stop. When I attach the same switch to the first switch that that run is plugged into, everything's fine again.

Argh! I hate networking...

Sounds like the 200' run was recently damaged somewhere. If nothing else was changed recently, it used to work and the switches are okay if they're next to each other. Inspect the cable run if you can or find somebody with a decent cable tester.

When you use the laptop, can you find out whether its running at half or full duplex?

I'm guessing the switch is doing something like running full duplex over a damaged cable or one side is half and the other is full. The laptop is being better about it and probably forcing itself to half duplex.

It should still be 30/5, or close. You definitely shouldn't see a dramatic dropoff. That, combined with the fact that putting S2 right next to S1 works fine, points to that cable being bad. Can you run a new cable? Maybe just along the hallway temporarily to see if it helps matters?

There is actually a limit for how many switches you can chain together like this but you're not hitting it.

Actually, I've found that the 30/5 test was a fluke. I think we're supposed to get and are getting 20/5. That sounds better anyway, and is a more common speed among ISPs around here.

I think on most attached computers I'm getting 18-19Mbps.

Unfortunately I don't have anything long enough to run it from switch 1-2. I suppose it's possible that the long run is damaged in some way, but this would seem unlikely. It's housed in a brand new underground conduit running from the new building construction to the old building, and there hasn't been construction going for months, so I can't see how anyone would have cut it or anything.

Is there a way to see if I'm at half or full duplex? I'm not sure how to tell. Isn't it on automatic by default? I suppose I could force it to full and see if it works or not.

Try forcing it to half and then full. It depends on what NIC you've got on where it does this.

Water damage maybe?

Do you know what is running in the conduit with your cable? Or even in adjacent conduits. It doesn't have to be physical damage you are looking for. Has anything changed on site recently, anything new electrical installed?

A quick and easy test for this is to use a longwave radio, if you are picking up massive mains humm across the range you have some kind of interference onsite that could jump over into your cat5 wiring over a distance like that.

Just an idea, and is a simple enough check if you do have a battery long wave set to try it out with.

Cheers

Cris.

Is there a chance that the wire pairs on that 200 foot run aren't the proper pairings?

Ethernet cables are designed with certain wire pairs going to certain pins, so that the "twists" in the wire pairs correctly do all of the necessary noise reduction for long cable runs. Someone crimping down an RJ45 plug on either end of such a cable run might not necessarily know about that little detail, and might have done it wrong.

If so, you might have symptoms like those you describe: Depending on what piece of gear you've got hooked up at either end of the cable, it might work fine or it might succumb to noise issues.

Tony: if it were improper pairings, that wouldn't explain why it stopped working when it once was.

Cris: nothing has changed on the site in a few months, and things were working fine until last week. We've called the contractor who installed the line, but they won't be able to get out here with a tester until tomorrow.


I just tried setting my NIC to full duplex manually, and it worked just fine.

It would explain it if there was an unexpected addition of new RF interference somewhere nearby. Hypothetical example: The air conditioning in the building starts getting used in the spring, and this is causing new electrical noise that wasn't there until recently.

If I'm understanding your problem description correctly, you're not seeing an obvious malfunction, rather, you're seeing an unexpected performance drop. Right? Bad wiring can do that at unpredictable times.




If you force your NIC to half duplex, does the performance drop to match what the other downstream computers are experiencing?

Oops. Didn't try the half duplex. I'll try it tomorrow.

I had to leave eventually, and the cabling contractor who put the conduit in will be coming by tomorrow, so hopefully we'll all be able to figure it out. Sheesh!

Half duplex works just fine.

Is it odd and/or telling that upload speeds remain constant in all configurations?

I'm still waiting for the cabling contractor.

Half-duplex tells you which pair of wires has been damaged. Probably a nick in the sheath, and some water in the conduit.

Or not.

I'm sorry, I wasn't being clear enough. I meant that when plugging my laptop in to the end of the 200' run it works fine on bot half and full duplex. Nothing still works if a switch is put on the end of that run.

I finally tested the one thing I hadn't before. I basically removed switch #1 from the equation, and put a little four port switch between the router and that 200' run. Oddly, the speed improved, but only to about 5Mbps down, and the upload dropped for the first time to something like 300Kbps.