Massive Packet Loss
Results 1 to 14 of 14

Thread: Massive Packet Loss

  1. #1
    Registered User
    Join Date
    Nov 2000
    Location
    Fountain Valley, CA
    Posts
    507

    Post Massive Packet Loss

    I'm supporting a 4 computer network: 3 have 98, one has 2000 Pro. It is a peer to peer network.

    Computer MAIN, a 98 machine, has the quickbooks data files on it, and the other three access quickbooks over the network.

    CLIENT1 and CLIENT2 are 98 machines. They don't have any problems accessing the quickbooks data files.

    CLIENT3 is the 2k machine. For the most part, it works flawlessly. But recently, their QB data file has been getting corrupted very frequently, and they have to restore the backup file. They called me out to check on the network, because something seemed fishy.

    I ran a "ping test" on all three CLIENT computers. I ran four instances on each computer like so:
    "ping main -l 65500 -t"
    so that each machine had four sessions open, pinging MAIN with 65500 byte packets.

    CLIENT1 and CLIENT2, as suspected, pinged just fine, with only about 3-5% packet loss over about 15 minutes.

    CLIENT3 (win2k box), pinged fine for about 5 minutes, but then took a major downswing, losing long strings of packets. After about 15 minutes, the loss rate reached nearly to 80%.

    Here is what I have tried:
    1. Replace NIC
    2. Replace hub with 10/100 switch
    3. Replace network cable

    None of these provided any help.

    I am currently downloading the Win2k service pack 2.

    Does anyone have any ideas what could be causing this phenomenon?

  2. #2
    Registered User
    Join Date
    Jan 2001
    Location
    Tulsa, OK
    Posts
    35

    Post

    Is there anything enviormental happening buy this one machine. Are the calbes striat into the hub or wall plates? Patch panel? if so plug one of the 98 machines cables to the 2k machine to make sure all the calbes and conections are ok . X
    No really That paper thingy you took out of the box with all the words on it was not packing material its called a "MANUAL"

  3. #3
    Registered User
    Join Date
    Nov 2000
    Location
    New Jersey
    Posts
    211

    Post

    Definitely check the environment. Also is TCP/IP the only protocol running?? Anything else running on the Win2000 machine in the background?

    I don't know from what I've seen the Windows 2000 TCP/IP stack blows away Micro$oft's previous o/s's. Let us know what you find.
    "Beer is proof that God loves us and wants us to be happy" -Benjamin Franklin

  4. #4
    Senior Member condor's Avatar
    Join Date
    Jan 2001
    Posts
    878

    Post

    if you are using a "Realtek" based NIC throw this POS out..

    try to use 10Mbit/ Half duplex..

    see if the problems persists..

  5. #5
    Registered User Higg's Avatar
    Join Date
    Jul 1999
    Posts
    295

    Post

    only for troubleshooting: try a direct connection with a crossovercable and ping the W2K machine

  6. #6
    Registered User
    Join Date
    Nov 2000
    Location
    Fountain Valley, CA
    Posts
    507

    Post

    thanks for the suggestions.

    - I don't think there's an environment issue. It's a small, clean office, with no visible environment variables.

    - I was thinking about switching to half-duplex myself, so I will give that a try. All machines are currently running at full-duplex.

    - The NIC was a Linksys 10/100, and I replaced it with a 3Com 3c905c.

    - The computers are plugged into wall plates and from there to the switch. I tried bypassing the wall plate by running a long cable directly from the computer to the switch, to no avail.

    I'll keep you posted. In the meantime, more suggestions are appreciated. -tk

  7. #7
    Registered User
    Join Date
    May 2001
    Location
    Canada
    Posts
    24

    Post

    If you are still using the 3Com 3c905c, make sure that you have the latest drivers... We were having the same issues with the exact same card using AutoCad... copying the file from machine to machine using Explorer caused no problem, however, if we attempted to open an autocad drawing across the LAN the file would become corrupted and useless.
    _ _ _ _ ____ _ _ _ _
    James Sprott
    [email protected]
    http://webscaping.net

    "Never put all your ranking officers in the same shuttle craft!"

  8. #8
    Registered User
    Join Date
    Nov 2000
    Location
    Fountain Valley, CA
    Posts
    507

    Post

    This problem has taken 5 hours of my life and I want them back.

    Yesterday I tried setting the computers to 10Mb/Half Duplex, to see if I could find anything new. However, in order to run my little ping test, I had to back down the packet size from 65K to 16K, otherwise, all the packets would be lost.

    At 16K packets, and at 10Mb/halfduplex, pretty much the same thing seemed to be happening, though not as severe. Instead of losing 80% of the packets, I was only losing 40% or so on the 2000 machine. The 98 machines still ran fine, at about 3-4% loss.

    I tried removing the NetBEUI protocol from the 2000 machine, and the 40% dropped to about 20%. I removed NetBEUI from the other computers, and it dropped to 1%, along with all the other computers. DUH! I thought. It was netBEUI all along!

    BUT.....

    That wasn't the end of it.

    Since I figured I had figured out the problem, I thought I would set the computers back to 100Mb, full duplex. Well, after I did that, I ran my ping test again, but ran it at 65K packet size, and there was my old 80% packet loss again.

    what the hell.

    At this point, I'm not really sure if the ping test is a great barometer to measure the network stability of these machines. I told the customer to go ahead and try running quickbooks again, and see if it crashes any more. It *SEEMED* like it was working a bit faster when I left, but I guess only time will tell. I just told him to backup about 3 or 4 times a day, so you don't lose much data if it happens again.

  9. #9
    Registered User
    Join Date
    Nov 2000
    Location
    New Jersey
    Posts
    211

    Post

    Hey you said you switched the hub to a switch, what kind of switch are you using??

    Personal opinion here: Look into using Intel Pr0 100+ cards, I don't care what anyone says they beat the hell out of any other NIC card. And I've seen quite a few at this point.
    "Beer is proof that God loves us and wants us to be happy" -Benjamin Franklin

  10. #10
    Registered User Higg's Avatar
    Join Date
    Jul 1999
    Posts
    295

    Post

    Originally posted by tk421:
    ... Well, after I did that, I ran my ping test again, but ran it at 65K packet size, and there was my old 80% packet loss again...
    Can you give me a sketch why you want to use 65k ping? Even if it's possible ... the standard Ethernets packet size is smaller 1518Bytes (including Header) and in normal networks the router will just drop your packets or split them despite your order!

    I'd suppose to try the standard ping-size of 100Bytes (upto king-size of 1500Bytes) ... if that works well -don#t bother about sending 65k-size unfragmented packets.

  11. #11
    Registered User
    Join Date
    Nov 2000
    Location
    Fountain Valley, CA
    Posts
    507

    Post

    The reason I am pinging with 65K is because I want to put stress on the network and see where it is breaking. If the 65K packets were getting lost on ALL the machines, then obviously, I would have thought to myself, this network cannot handle 65K packets. HOWEVER, since 3 machines handled the 65K packets just fine, dropping only 3-5% of them, while the 2K machine was dropping them 80% of the time, I felt that the 65K packet test was successful in illuminating a weakpoint on my network.

    Do you not agree?

    I completely agree that in normal network use, 65K packets will never be seen. Pinging with 65K pacekts was a *stress* test.

  12. #12
    Registered User Higg's Avatar
    Join Date
    Jul 1999
    Posts
    295

    Post

    Problem is that you'll never know if it's right or wrong to drop 65k packets... let's say, if it worked once, it should work twice... but if it never worked, I wouldn't worry about - I'd change the test...

    Perhaps you've done stress to the packeting and segmenting software on the machine itself...

  13. #13
    Registered User
    Join Date
    Jul 2000
    Location
    Huntington Beach, CA, USA
    Posts
    1,515

    Post

    Are the cards and the switch both set to 100/Full? If the switch is set to auto you could have a speed/duplex negotiation problem (have seen it many times before).

    Try setting the switch to 100/full.
    Death is lighter than a feather - duty heavier than a mountian.

    The answer to your question is: 00110100 00110010

  14. #14
    Registered User
    Join Date
    Nov 2000
    Location
    Fountain Valley, CA
    Posts
    507

    Post

    well, if the network was working fine, I wouldn't care if it couldn't ping 65K. But since the network is failing, and predominantly on the 2000 machine, and it is the only one that has trouble pinging at 65K, then I don't think I am crazy in suspecting a correlation.

    Thanks cyber, I would try that, but the switch does not have any management functions. Just a little 8port linksys.

Bookmarks

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •