[mpls-linux-general] a quesion about ldp-portable fail :(

[Folke A. <aeo...@ho...>]

hi James,

  i am using mpls-1.172 and ldp-portable-0.305.
my network topology is very simple :
    +-------+1.1      +---------+2.1
  --+ cisco +---------+ linux 1 +---+
0.5+-------+      1.2+---------+   |
                                    |
                +---------+ 2.2     |
           -----+ linux 2 +---------+
             3.1+---------+

the network adaptor is 10Mbps ether link

  between linux1 and cisco , i use
ldp to distribute label.
  between linux1 and linux2 i setup
a static lsp use mplsadm2.

on linux1 :
eth0    0       14
eth1    0       11
0x02718001 0/0/0 gen 10008 0 1 POP PEEK
0x02719001 80/4934/0 gen 10009 0 1 POP PEEK
0x0271a001 0/0/0 gen 10010 0 1 POP FWD(0x00000002)
0x075f8001 51/2952/0 gen 30200 0 1 POP FWD(0x00000006)
0x00000002 16552/15440008/0 7
    EXP2TC( EXP(2)->TC(0050) EXP(3)->TC(0048)
    EXP(4)->TC(0030) EXP(5)->TC(0028)
    EXP(6)->TC(00b8) EXP(7)->TC(00fc) )
    PUSH(gen 230) SET(eth1,11.0.2.2)
0x00000006 258/12285/0 3
    EXP2TC( EXP(2)->TC(0050) EXP(3)->TC(0048)
    EXP(4)->TC(0030) EXP(5)->TC(0028)
    EXP(6)->TC(00b8) EXP(7)->TC(00fc) )
    PUSH(gen 3) SET(eth0,11.0.1.1)
mpls03  0x00000002
mpls30  0x00000006

note that the cross-connect "FWD(0x00000002)"
and the "FWD(0x00000006)"
is created manually using mplsadm2 after i
get the incoming label 10010 through ldp.
(by the way: can i use ldp to automatically
bind a incoming label to a static out-label,
if i need to do so, what should i do?)

well,
my question is , if i run a traffic generator
like 'tcpblast' as the following on LINUX 2 :
tcpblast -b 16KB -c 1M -R 2MBps --udp 11.0.0.5
then i got a congestion on my network , and
after the tcpblast keep running about 50
seconds or 1 minute, the label 10008, 10009
and 10010 are both disappeared. and i
telnet to the vty, and try:
show ldp neightbour
   i got 'no ldp peer' message or sth. like that.
show ldp session
   i got nothing too.
show mpls database
   still nothing delight me. :(

if i shutdown the tcpblast , i will get some
new labels, such as, 10001, 10002...

at first i suspect that it is because the
congestion made the ldp daemon cannot send the
ldp hello messages correctly. but i setup
qdiscs using tc and us iptables to mark the
local generation packet as EF packets ,
thus , i can see one packet send out from the
EF queue about every three seconds -- the
hello interval i use, (my keepailve interval
is 59 seconds) also , i can catch one hello
packet from cisco every 5 seconds during the
5 minutes of testing.

i am not use the ehtereal-analyzer or some
other packets-catching tools. and my statistic
figure is the output of 'iptables' and the 'tc'

i will use the ehtereal-analyzer to make
sure that during the congestion there do have
or haven't hello messages or keep-alives
messages sometime later.

  if there do have these messages, there
may be some bug in ldp daemon.

would you please have comment on this ?
thanks :)

folke



_________________________________________________________________
Add photos to your messages with MSN 8. Get 2 months FREE*. 
http://join.msn.com/?page=features/featuredemail