Network Connection Issues

[reggie14]

I've been having a problem with my network connection seemingly dropping about once a day. Its actually a bit more complicated than that...

Roughly once a day, usually while I am streaming SageTV on my client, I will stop being able to browse the internet, or connect to other computers on my internal network. I can't even ping other internal computers. However, the SageTV video stream continues to work for about a minute (not from a buffer, there is network traffic), after which time Sage crashes (presumably because for some reason it lost its connection too). Right after that point, everything starts working again.

I should note that stopping SageTV myself before it crashes does not fix the problem. And, no matter what, the problem will alwys resolve itself after about 1-2 minutes, whether Sage is running or not.

I don't mean to be claiming that Sage is at fault for this issue. But, I'd really like some help troubleshooting this. I thought it was my router/switch at first, but I have the same issues with other switches. I suppose it could be the network adapter on my motherboard, but then why would Sage's video stream continue working for a period of time.

Perhaps its a software/driver issue (note, my MB drivers are up to date), but I really don't want to go through the reformat/reinstall process unless I have to.

Has anyone else had any problems similar to this?

[Para]

Originally Posted by reggie14

I thought it was my router/switch at first, but I have the same issues with other switches.

Just to be certain. Did you check the DHCP lease time of your router? Is it set to greater than 24 hours (be aware that some routers list lease time in minutes)

[reggie14]

Thanks for the reply.

My DHCP lease time is set at 24 hours.

While I guess I'm not certain, I don't think this is a DHCP issue. I've had a problem with DHCP leases expiring and not properly renewing before, and I would always get an error message showing up in my Event Viewer. After changing firmware on my router I got rid of that problem.

When this problem occurs, I don't get any warnings or errors in the Event Viewer (except for something about Sage crashing, if it was running).

[gplasky]

You're saying you're having this problem roughly once a day and your DHCP lease is set for 24 hours. Hmmmmm. Why don't you increase your lease to 48 hours and see if you have the issue once every 2 days. Just because you don't get an error message doesn't mean it's not a DHCP issue happening.

Gerry

[reggie14]

Thanks for your help. I've really been racking my brain trying to figure out what could be causing this problem, so its nice to get some fresh ideas. I will try the advice about changing lease time. I should point out a couple things though. First of all, I stress the roughly once a day. Its by no means consistant, and its certainly not at the same time each day (although, as you know, you renew your DHCP leases well before they expire).

Also, the thing that really confuses me is that the SageTV client connection to the sage server continues to function properly for a period of time. I can see I'm getting data over the network because the switch activity lights are flashing, as is my network adapter led, and windows reports network activity. Why should that continue working while other things do not?

The only thing I can possibly think of is that somehow the TCP-IP stack on my computer has gotten screwed up, and sometimes it refuses to create connections. Existing connections like SageTV, AIM, MSN, continue to work, but new ones, like one created to a webserver, will not work.

Of course, what somewhat screws up this theory is the fact that I can't send UDP traffic (ie, ping) either. It seems like if I can send traffic over an existing TCP connection I should be able to send UDP traffic, since that doesn't need to create a connection.

[mike1961]

Reggie - I sell a relational database management program to businesses which also has a lot of "network traffic" and I can tell you for a fact that 99% of the time when my clients have problems it is due to a "network connectivity" problem. This means that when there is "high traffic" on the network you can get errors. Also, even when there is low traffic, errors can occur. You want to be sure you do not have this problem. Here's how to isolate it:

1. Over 95% of the time it is due to a bad connector on the network cable. Consider replacing the cable from the workstation with the problem to the router. Of course it could be a faulty cable from the server to the workstation. You can "test" the cable as well but don't use a cable tester because it won't work. I can tell you how to test the cable if you want to know the details - otherwise, just replace it.

(EDITED: if it is a pain to replace - start by just getting a 100 foot test cable and run it from the workstation to the hub and warn family members not to trip over it. Test it for a day or two. Keep in mind you could also have a faulty cable from the router to the server. If it turns out to be a faulty cable - you still probably don't have to change the cable because chances are it is right at the connector so get an RJ45 crimper and connectors and just replace the connectors at the cable...but it could be a faulty cable as well if the rats are eating away at it in the attic or it's under a heavy desk or spiked mats, etc.)


2. (less likely) It could also be a faulty network card in the workstation or server or both.

3. (less likely) it could also be a faulty connection at the hub and/or router. Try plugging the cable into a different jack in the router.

4. (less likely) could also be a bad powersupply or mobo or memory.

Again - 99% of the time, it is a faulty cable without question. And most of the time it is right at the connector. But, I have seen cables under chair mats with spikes in them, pushed under the carpet right under the nails that stick up to hold the carpet down and also under heavy objects like desks and file cabinets.

Start by suspecting a faulty cable or cables somewhere. Consultants I speak with all the time "play it down....no no it's not that I've tested it" then I tell them how to really test it and everything changes. Like my program - Sage is a "heavy network traffic" program and therefore you want your network in tip top shape.

Mike

[reggie14]

Mike-

You know, I actually thought that it could be a cable problem before, so I used a cable tester and it seemed to be ok. I didn't realize that those tests aren't very reliable. I'm sure I have another cat5e cable around here somewhere, so I'll give it a try.

Thanks for the heads up.

[mike1961]

Cable testers are not reliable at all. They will give you a lot of "false negatives." In other words, yes, when the cable is not transmitting and is dead, they show that. But, when it works and is faulty, they don't catch it.

Now - here's what I discovered as an excellent way to test a network Cat5 or any network cable (much better than a cable tester):

1. Create a batch file in DOS. I call this file dodir.bat with the following information:

DIR F:
DODIR

Assuming F is the network drive, the batch file runs indefinitely. The DIR command is not cached so this batch file will constantly be getting live info on the network. Make sure there are plenty of files in the F drive or network directory you are getting a directory of (like 50 or more).

2. The directory should scroll FAST on the screen and there should be no pauses anywhere (unless or course the file server is busy saving or copying something so be sure to do this when you do not expect any other major disk activity on the file server).

3. While the directory is scrolling fast on the workstation, place your finger on the network cable about an inch or two away from the connector at the workstation and push the cable UP at around 12 oclock for about 10 seconds. The directory should continue scrolling FAST with no pauses. Next, push it down at around the 6 oclock for 10 seconds and same thing at 9 oclock and 3 oclock positions. In all positions, the directory should scroll fast. If not, you have a problem right at the connector. If it works, try the same thing again at the connector at the router. Also, try it with the server's connectors while polling the workstation. Often, they are just not crimped down right. Sometimes you can even look at the RJ45 and you can see the gold pins are not crimped FLUSH down on the cable due to substandard crimpers.

4. This method mainly tests for bad connectors which is usually where the problem is. It could of course be the cable itself especially when they are under 200 pound desks and stuff. However, if you still have not found it, try plugging the cable in a different jack on your router.

5. I don't recall if you said this problem was on all your workstations or just one. But, I would test the cables on all the computers anyway. If the problem is on all the computers, then you could of course have several bad cables as well as a bad cable at the server.

6. It's usually the cable. Like I said though - other causes could be a bad router or network card. One time, I had to disable the onboard networking on the motherboard and just insert another card in a client's computer.

7. Once in a while it can of course also be software related like when you have spyware, viruses and of course the virus checkers as well (they really tend to do a number on your computer when not configured right). Make sure your virus checkers are not scanning your server but only the workstation - if you have any virus scanners. Also, if you do have a virus checker and it's norton - be sure you have the corporate edition.

Keep at it and let us know what you find.
Mike

[mike1961]

Cable testers are not reliable at all. They will give you a lot of "false negatives." In other words, yes, when the cable is not transmitting and is dead, they show that. But, when it works and is faulty, they don't catch it.

Now - here's what I discovered as an excellent way to test a network Cat5 or any network cable (much better than a cable tester):

1. Create a batch file in DOS. I call this file dodir.bat with the following information:

DIR F:
DODIR

Assuming F is the network drive, the batch file runs indefinitely. The DIR command is not cached so this batch file will constantly be getting live info on the network. Make sure there are plenty of files in the F drive or network directory you are getting a directory of (like 50 or more).

2. The directory should scroll FAST on the screen and there should be no pauses anywhere (unless or course the file server is busy saving or copying something so be sure to do this when you do not expect any other major disk activity on the file server).

3. While the directory is scrolling fast on the workstation, place your finger on the network cable about an inch or two away from the connector at the workstation and push the cable UP at around 12 oclock for about 10 seconds. The directory should continue scrolling FAST with no pauses. Next, push it down at around the 6 oclock for 10 seconds and same thing at 9 oclock and 3 oclock positions. In all positions, the directory should scroll fast. If not, you have a problem right at the connector. If it works, try the same thing again at the connector at the router. Also, try it with the server's connectors while polling the workstation. Often, they are just not crimped down right. Sometimes you can even look at the RJ45 and you can see the gold pins are not crimped FLUSH down on the cable due to substandard crimpers.

4. This method mainly tests for bad connectors which is usually where the problem is. It could of course be the cable itself especially when they are under 200 pound desks and stuff. However, if you still have not found it, try plugging the cable in a different jack on your router.

5. I don't recall if you said this problem was on all your workstations or just one. But, I would test the cables on all the computers anyway. If the problem is on all the computers, then you could of course have several bad cables as well as a bad cable at the server.

6. It's usually the cable. Like I said though - other causes could be a bad router or network card. One time, I had to disable the onboard networking on the motherboard and just insert another card in a client's computer.

7. Once in a while it can of course also be software related like when you have spyware, viruses and of course the virus checkers as well (they really tend to do a number on your computer when not configured right). Make sure your virus checkers are not scanning your server but only the workstation - if you have any virus scanners. Also, if you do have a virus checker and it's norton - be sure you have the corporate edition.

Keep at it and let us know what you find.
Mike

[Steerpike]

Since the previous poster went to such good lengths to talk about network cabling, I'd just like to add ... from bitter personal experience ... if you crimp your own RJ45's, read up on exactly how to do it, and do it right. Practice a few times, throw away your first attempts.

I was once up till 6am the night (morning) before a major presentation in a hotel with hundreds of attendees, troubleshooting network issues. It turned out the guy who crimped the cables assumed he could run the wires in logical color sequence (green/green stripe; orange/orange stripe; etc). These cables actually worked over short distances, but ... on some longer runs, they started to fail. The actual spec for cat 5 / RJ45 involves 'splitting' one of the color pairs. See this page: http://www.aptcommunications.com/ncode.htm
The actual colors don't matter, but what DOES matter is, the fact that (in the first diagram on the site) the green/green-stripe pair spreads around the blue/blue-stripe pair.

We finally cut off the ends and re-crimpted, and our problems went away.

[reggie14]

Well, I replaced the networking cable a few days ago and I haven't had any problems since. It appears that was the problem. Thanks a lot everyone for your help, especially you, Mike.

[mike1961]

Steerpike - Very good point. You are absolutely correct that there is a proper logical ordering to the colors and it's not the striped color followed by that color (ie: it's not orange strip, orange, blue stripe, blue, green stripe, green, etc.). Although that will work, I think it not only increases the errors but also decreases the speed down from 100 mgs or 1 g down to around 10 mbs.

Another important factor is how many cheap crimpers do not give the connector a good "bite" down on the conductors. You can often see this at the two end pins when looking at the side of the connector. Ideally they should be flush down with the cable.

Reggie - glad to hear you resolved the problem.

Mike

[Steerpike]

Not to hijack this thread, but ... the original problem is solved now, and this is such an obscure issue that people don't appreciate. It took me ages to figure out why the ordering of wires actually matters ... and while writing my last post, I realised I could not remember why!

The reason is, I believe, that the devices using the cable - eg, your hub, or network card - 'sends' on one pair, and 'receives' on another pair. All devices send and receive on pre-determined pins - obviously, otherwise devices couldn't communicate. There are 8 pins in total. the pins used to send data are pins 4 and 5 (the center pins). The pins used to receive are pins 3 and 6 (the pins either side of the center pins) ("send" and "receive" are relative and irrelevant to this). Thus, pins 1 and 2, and 7 and 8, are not generally used.

Since you are sending on pins 4 and 5, wires 4 and 5 need to be a twisted pair. The color coding in a cat5 cable (color, color-stripe) denotes a pair - thus, blue and blue-stripe are a twisted pair inside the outer cable. Thus, you need to ensure that one pair goes into connectors 4 and 5. The second pair (eg, green, green-stripe) must connect to connectors 3 and 6 for the 'receive' signal. If you don't do this, you still have perfect electrical connectivity between the relevant pins, but you don't get the 'send' and the 'receive' signals running over a twisted pair, and thus, over longer lengths, or in electrically noisy environments, your comms are susceptible to interference.

Finally, not only do you need to crimp well, you should not 'untwist' the pairs too much - untwisting the pairs defeats the important benefit that a 'twisted pair' offers. Joining two shorter lengths into one long lengh using a coupler is also a no-no for the same reason - the couplers 'break' the twisted pair continuity.

Not a great explanation, really needs a diagram, but ... hopefully of use to someone!

By the way - this does also illustrate the fact that pins 1,2 and 7,8 are totally up for grabs, and some adapters make use of this fact to run TWO ethernet circuits over one cable. Other adapters let you run power over these 'spare' lines.

[mike1961]

I agree with everything you said except that last part about "untwisting" the twisted pair. It's really not possible to untwist it since 99% of it will be insulated. Only about 1 inch on both sides of the cable will be "untwisted" to fit into the connector and I don't think that will be relevant. In fact, I always completely untwist the last inch so that it better fits into the connector and there is less "tug" when crimping everything down. Otherwise, you can have a situation where not all the wires are all the way in the connector (this was really a problem years ago with 4 conductor cable. 8 conductor RJ45 seems a bit easier to crimp and manage).

Edited - I find RJ45 easier to manage than 4 conductor (RJ22 and twisted pair) because it's longer so the wires go deeper into the connector.

Mike

Originally Posted by Steerpike

Finally, not only do you need to crimp well, you should not 'untwist' the pairs too much - untwisting the pairs defeats the important benefit that a 'twisted pair' offers. Joining two shorter lengths into one long lengh using a coupler is also a no-no for the same reason - the couplers 'break' the twisted pair continuity.

[Menehune]

Originally Posted by mike1961

Only about 1 inch on both sides of the cable will be "untwisted" to fit into the connector and I don't think that will be relevant.

I disagree. Ethernet specifications only allow a maximum of 1" of untwisted pairs over the entire cable length-only one-half inch per connector. Leave the pairs twisted to preserve the speed rating of the cable and reduce crosstalk. If you need can't push the outer insulation under the strain relief into the connector, you are probably using the wrong type of connector-i.e. forcing a "round" cable into "flat" connector.

What I do is remove about three inches of the outer insulation from the cable and untwist the conductors. I then flatten the conductors iinto the correct color sequence and so they can be fed into the RJ45 connector. I then trim off the excess leads perpendicular to the cable (so they bottom fully into the connector) leaving 1/4 to 1/2 inch of untwisted conductors. There are also "pass thru" RJ45 connectors which allow the conductors to pass thru the front face of the connector so the outer insulation can fully seat into the connector body.

I agree, CatV solid is much easier to terminate than stranded "satin" phone cable

[mike1961]

I see your point....that is basically what I do too. I strip out about 2 inches or so which makes it easier to "flatten" the wires then cut them to the point where they will fit all the way in the connector and some of the insulation goes in as well so you are right that there is probably no more than around 1/2 inch of untwisted wire.

Mike

[MrApollinax]

Originally Posted by Menehune

I agree, CatV solid is much easier to terminate than stranded "satin" phone cable

I would also like to stress the importance of getting the correct RJ45 connectors if you are going to make your own cables. If you have solid wires, get connectors that are designed for solid wire, if you have stranded get stranded wire connectors. My personal experience also tells me to STAY AWAY from connectors that are advertised to support both.

[Steerpike]

Since there has been some interest in this topic of cat 5 and RJ45, would anyone have a strong recommendation for a specific brand of cable and connectors, not to mention a crimper? I actually need to buy the whole lot, so a tried-and-true recommendation would be nice!

I'm specifically interested in 'moderately flexible' cable, not the stiff stuff that we ran through above the ceiling tiles last time I did this!

Thanks.

[elaw]

I'll take the connector discussion a step further and offer this 3-step procedure for making good patch cables with solid wire: 1) throw away the solid wire, 2) get some stranded wire and proper connectors, 3) crimp with good crimper.

I've been in the IT business for a long time and I really try to avoid homemade patch cords. Most professionally-made ones are pretty cheap considering the quality you get (if they're not cheap, go to a different store). But I've had absolutely terrible experience with ones made from solid wire. When the RJ-series of connectors was originally designed, they were not designed to provide the contact force that's needed for a reliable connection with solid wire. I know some connectors are supposed to work with it, but I'll estimate that 75% of the solid-wired patch cords I've ever encountered have caused problems.

Just to be completely clear, my comments pertain only to cables with plugs on the ends. Jacks that are made for premises wiring work just fine with solid wire, that's what they're designed for.

Eric

[mike1961]

Agreed that when you are going 10-20 feet, one may as well just get a professionally built cable (even though they too can be defective from time to time). But, when you have taken the time to run a cable perhaps a hundred feet through walls and crevices then a) you have to make the holes in walls a lot bigger if you don't want to attach your own connector b) if the connector is bad, it's easier to just crimp on another connector then rewire a new cable.

Mike