Need help with losing connectivity with device

[TiggerDaKat]

We have a system with 5 Windows 2000 servers and one AS/400 server. WE lose connectivity with the AS/400 quite frequently. When we ping it, we usually get 1 reply and the rest time out. pinging all the other servers works 100%. The guys from IBM said there are too many collisions on our LAN. Why would all other servers work fine then?
We've purchased and installed 3 new 3com Superstack 3 24 port switches, thinking that might help the problem...but it didn't. We had IBM replace the NIC in the AS/400 and that didn't help either.
Where should I go from there?
Thanks in advance.

Eric

 

[hitender]

Tigger,

First of all check the cable connectivity to AS server.Some times in these type of problem we change NIC SWITCH but don't take care of cable laid to whihc server is connected.
Get the cable checked.Because u have checked at Layer 2 so chacke with layer 1(physical layer.)i.e cable Hitender (BE.CCNA)
Best Of luck
Try & Try Till U succedd and update others on u r achievments

hit_singh@rediffmail.com

 

[Accessdabbler]

I don't know how your network is set up but I've had a problem with a router that would work fine for awhile then start acting goofy. Pinging systems on the other side of the router would result in only 1 or 2 successful attempts. Rebooting the router would clear this problem immediately.

That said, I think you need to test the cable to the AS/400 server first though.

 

[jimbopalmer]

Minor aside: Years ago I had a network with one DEC VAX and about 10 suns, for resons I never understood, when a collision occured the random time out picked by the VAX was about 1000 times as long as the random timeout picked by the Suns. When I did backups at night, the VAX was unresponsive, but it worked fine durring the day. (don't you hate problems that only happen while you sleep?)

Subnetting the VAX into it's own subnet solved the problem, as we did not own switches back then. Anyway I HAVE had the collison problem affect one vendor's NIC more than another vendor's. I tried to remain child-like, all I acheived was childish.

 

[TiggerDaKat]

Some more information for you guys...
We did change out the cable. That didn't do the trick.
The ip address of the AS/400 is 172.16.0.1 subnet mask of 255.255.0.0.
All of the PC's are on the same subnet with the ip numbers of 172.16.10.50 through 172.16.10.100 with sybnet masks of 255.255.0.0.
We aren't using any routers between them. I was told this is a acceptable way to address all devices.
Does anyone see any problems with this?
Thanks again in advance.

 

[Accessdabbler]

I don't see a problem with the addresses but I think you need to test the cable that's already installed. You may have interference somewhere along the route.

 

[TiggerDaKat]

We had IBM run a trace on the AS400 and they told us there was packet loss on the network due to excessive collisions. Is there any software or hardware out there that will tell us where the source of the collisions are?

 

[jimbopalmer]

If you still have excessive collisions after you installed the switches, I am going to suspect duplex mismatch problems.

Try forcing your IBM NIC to 100/half duplex or 10/half duplex.
3com superstack 3's can be managed or unmanaged, if it is managed, try forcing the duplex at the switch as well.

(Half duplex would, in theory, have MORE collisions than full duplex, but a situation where one side is half duplex and the other full duplex has many more than all half)

Computer NIC Switch port connection
Auto Auto Usually works fine
Auto Full (managed) Fails with collisions,slow
Auto Half (managed) Works
Full Auto Fails with collisions,slow
Full Full (managed) Works
Full Half (managed) Fails with collisions,slow
Half Auto Works
Half Full (managed) Fails with collisions,slow
Half Half (managed) Works

Auto to Auto fails if the manufacturers read the standard differently, Cisco and Apple are usual suspects at reading standards their own way.

Notice that if your 3com is unmanaged we have fewer choices, and that Half to Auto is our best shot if the IBM NIC was on Auto and not doing well. (if someone put the IBM NIC on Full, put it back on Auto, and try that) I tried to remain child-like, all I acheived was childish.

 

[hitender]

hi
There is some EME interference inpath of u r server cable.Do check for that.

If u have device Called PENTASCANNER then test with that. Hitender (BE.CCNA)
Best Of luck
Try & Try Till U succedd and update others on u r achievments

hit_singh@rediffmail.com

 

[juggy]

IBM... For you sanity. If every time you ping the device it comes back reqest timed out and sometimes connects you could have a traffic related problem or the AS400 network card could be configured wrongly its not comming back to me but I had a similar problem with a 3745 and its because the frame size was setup for a line and not for an ethernet card thus too much traffic the box stopped responding. I know you can check this on an as400 from the menu just cant remeber the exact one if in doubt call IBM and get them to confirm the packet size. Guys help out with a good packet size for an ethernet card my mind is blank at the moment it shouldnt be lower that 512 though cause its a runts. If this looks good then one night when you get some down time take the as400 and get 1 laptop or your pc only and put them on the same hub first. Ping. Does it work correctly. Yes? Traffic setup related issue IBM (as long as the frame size isnt too small) were right. If it still doesnt work then tell IBM what you have done and tell them to get there arxxs into gear and sort out your AS400. If it does work fine try pinging with a few differnet parameters to see if it still handles the load ok ie ping -l and put a big number here thats legal. IF this still looks ok then the problems are with your lan and traffic try disconnecting users off if they connect via a different hub.. Another one for IBM. If you use a switch each port is a collison domain therefore the AS400 in a port alone shoudnt experience collison problems so duplex settings (As advised above) would be a good start. A good rule ive always used is start with little ie one switch only.. Still have the problem remove devices 1 x 1 until its sorted if it still isnt sorted switch is configured wrong/badly. I know for a fact AS400 handle TCP/IP fine with only 1 card they are bloody good. Let us know how you get on.

 

[TiggerDaKat]

We finally found the problem.
There was a router attached to the system which had an incorrect subet mask set to 255.255.255.0. It should have been 255.255.0.0
This screwed up the ARP cache. When we fixed it, we still had to reset all the devices which cleared the cache.
Everything worked well from that point on.
What's strange is that it ONLY affected the pinging to the AS400. None of the Win2K servers had the same symptoms.
Thanks for everyone's help. It was quite a learning experience.

Eric