[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference netcad::hub_mgnt

Title:DEChub/HUBwatch/PROBEwatch CONFERENCE
Notice:Firmware -2, Doc -3, Power -4, HW kits -5, firm load -6&7
Moderator:NETCAD::COLELLADT
Created:Wed Nov 13 1991
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:4455
Total number of notes:16761

4374.0. "problem transferring 'large files' across DECswitches" by NNTPD::"[email protected]" (Feynman Lo) Wed Apr 23 1997 16:51

(Urgent...)

We just installed a network project on customer and encountered
a strange problem. The network is composed of  DECswitch
900EFs, 900ETs, 900EFs on two DEChub900s.

Simplified diagram:

 
                        ( 10.18.0.0 ) -  HP Unix
                            |
                      ( Cisco network , OSPF)
                        /            \
    10.91.0.0     (Token ring 1)      (Token ring 2)   10.92.0.0
                      |                   |          \      
                      |                   |            \      
          10.91.91.44 |                   |10.92.92.45   \
                   900ET                900ET              \       
        10.170.170.46 |                  |10.170.170.48      \
                      |                  |                    \
                      |                  |                     \
                      |                 900EF                   \
                      |                  /                       \
                      |       ======= FDDI ==== 10.170.0.0        |
              Ethernet|                |                          |Token ring
                      |                |                          |
                      |              10.170.170.56                |
                       ---------[Digital Unix]---------------------
                  10.91.91.40                              10.92.92.40
                       

All routing reachibility seems no problem. So we don't think
it is a routing problem.

The Digital Unix(3.2G) has a route to 10.18.0.0 via 10.170.170.48.
The HP Unix will go to 10.170.170.56 via 10.92.92.45 path
The HP Unix will go to 10.92.92.40 via Token ring 2 directly
The HP Unix will go to 10.91.91.40 via Token ring 1 directly

The FDDI is formed by 900EFs and 900FHs on HUB900 backplane

When we telnet to 10.170.170.56 and do FTP to get a file of
e.g. 10K size from HP, or do a 'ls -l' of long directory from HP,
it hangs and fails. But doing a small file transfer (e.g. 100 bytes)
has no problem.

When we telnet to  10.91.91.40 and do the same, we got the same result.

When we telnet to 10.92.92.40, this symptom does not occur.

When the transfer direction is reversed, e.g. putting files to
HP, we got no problem.


I suspected it is related to IP fragmentation (enabled on 900EF)
but it seems no help when I tried to disable it.

What is the cause? I realise we got a critical problem and 
we are very serious to seek for help. Any idea
will be vey much appreciated.

Regards,
- feynman

[Posted by WWW Notes gateway]
T.RTitleUserPersonal
Name
DateLines
4374.1Correction to the diagramNNTPD::"[email protected]"Feynman LoWed Apr 23 1997 17:0032
Sorry I made some mistakes to the 900ET on the Token ring 1 side.

Note that all hub modules are on the two DEChub900s
This is the correction.

Simplified diagram:

 
                        ( 10.18.0.0 ) -  HP Unix
                            |
                      ( Cisco network , OSPF)
                        /            \
    10.91.0.0     (Token ring 1)      (Token ring 2)   10.92.0.0
                      |                   |          \      
                      |                   |            \      
                      |                   |10.92.92.45   \
   (switch-only mode) 900ET                900ET              \       
                      |                  |10.170.170.48      \
                      |                  |                    \
                      |                  |Ethernet                     \
                      |                 900EF                   \
                      |                  /                       \
                      |       ======= FDDI ==== 10.170.0.0        |
              Ethernet|                |                          |Token ring
                      |                |                          |
                      |              10.170.170.56                |
                       ---------[Digital Unix]---------------------
                  10.91.91.40                              10.92.92.40
                       


[Posted by WWW Notes gateway]
4374.2NETCAD::DOODYMichael DoodyThu Apr 24 1997 16:318
I doubt IP fragmentation is the problem, but anyways
turning it off at the switch would cause more problems.

My guess would be that you have a problem with the
adapter/driver/ IP config setup. You might want to look in the 
digital_unix conference.

-Mike
4374.3ftp problem solved adjusting MTUNNTPD::"[email protected]"Feynman LoSun Apr 27 1997 04:3833
I haved the ftp problem at the 10.170 (FDDI) side by
adjusting the MTU size of the 900ET routers to 4399
which is the one comparable with custoemr's token ring
network.


By at the 900ET with switch-only mode side, the problem
is even worse. We fail to ping the Ethernet interface
of Digital Unix connecting to the 900ET switch, when
we do this from an subnet across a  router, no matter
Cisco or our 900ET routers. We can succeed to do so only 
after I ping the router interface from the Digital Unix
side. But after a while, the problem re-apperar again.
The ftp problem still persist, despite the MTU of the
switch has been adjusted. 

This occurs even when we do it on a PC  connected to
a 900EF switch on the 10.170 side. In this case , the
path only involve : 

  PC-Ethernect -> switch by 900EF to -> router-mode 900ET  Ethernet
  -> route to router-mode 900ET Token ring 
  -> switch-only mode 900ET Token ring
  -> switch to switch-only mode 900ET Ethernet
  -> Digital Unix Ethernet 


It sounds like that the 900ETs in switch-only mode fails
when connected to the token-ring routers. 

-feynman

[Posted by WWW Notes gateway]
4374.4bridged packets not fragmentedFORBIN::WILKINSONSun Apr 27 1997 17:0112
    I am not sure of my product names, but if the 900ET is the
    one of the token ring to ethernet switches based upon the Comet
    software code base, then it is very likely that IP fragmentation
    does not exist for bridged packets.
    
    Originally, this code base only provided IP fragmentation for
    IP packets that were routed.  We extended the functionality for
    the VNswitches so that bridged packets are also fragmented.
    Check with product management on when this functionality will
    be merged to other products.
    
    Hugh
4374.5So...then hosts have to support Path MTU discovery?NNTPD::"[email protected]"Feynman LoMon Apr 28 1997 06:5719
Thanks for indication.

If the 900Et switch-only mode does not support IP fragmentation
(like the 900EF does), can I conclude that we can only make
it works switching between Token ring and Ethernet  by either:

1) The hosts support Path MTU discovery. (Is it true for Digital Unix?)

or 

2) Making the Token ring network to have the same MTU as
   Ethernet ?? (I think it is impossible in the customer's case)

If both cases cannot be done, does it make sense to use
switch-only mode in 900ET any more?

-feynman

[Posted by WWW Notes gateway]