[c-nsp] Huge RTT on ATM, is there a hidden queue somewhere?

Daniele Orlandi daniele at orlandi.com
Tue Oct 28 06:24:24 EDT 2008


Hello,

I'm experiencing weird delays on ATM VCs that I'm unable to troubleshoot.

I am saturating the VC by downloading a large file (via scp) between two hosts 
directly connected to the routers[0] under test.

I reduced both the tx-ring-limit and vc-hold-queue to the minimum allowed:

interface ATM2/0.7 point-to-point                                    
 description ULI-SEVESO-125473/44                                    
 mtu 1500                                                            
 ip unnumbered Loopback0                                             
 ...
 pvc uli-seveso-hdsl 200/35                                          
  ubr 1000                                                           
  tx-ring-limit 3                                                    
  vc-hold-queue 4                                                    
  oam-pvc manage                                                     
  encapsulation aal5snap                                             
 !      

The output queue is empty:

gw-dsl#show queueing int atm2/0.7
  Interface ATM2/0.7 VC 200/35
  Queueing strategy: fifo
  Output queue 0/4, 0 drops per VC

I don't know how to check the tx-ring usage per-vc.

However, i see RTT go up to 700-1000 ms with an average of 600ms:

64 bytes from vega.uli.it (62.212.0.2): icmp_seq=345 ttl=60 time=639 ms 
64 bytes from vega.uli.it (62.212.0.2): icmp_seq=346 ttl=60 time=654 ms 
64 bytes from vega.uli.it (62.212.0.2): icmp_seq=347 ttl=60 time=653 ms 
64 bytes from vega.uli.it (62.212.0.2): icmp_seq=348 ttl=60 time=535 ms 
64 bytes from vega.uli.it (62.212.0.2): icmp_seq=349 ttl=60 time=680 ms 
64 bytes from vega.uli.it (62.212.0.2): icmp_seq=350 ttl=60 time=708 ms 
64 bytes from vega.uli.it (62.212.0.2): icmp_seq=351 ttl=60 time=723 ms 
64 bytes from vega.uli.it (62.212.0.2): icmp_seq=352 ttl=60 time=716 ms 

AFAIK the ATM network should not introduce huge delays. I don't know how OAM 
cells get treated, but under load I see the loop cells come back after 20-30ms 
so it shouldn't be a ATM-newtork related delay:

Oct 28 11:39:25.404: ATM OAM LOOP(ATM2/0.7) O: VCD#7 VC 200/35 CTag:0x3F9B9                                       
Oct 28 11:39:25.432: ATM OAM LOOP(ATM2/0) I: VCD#7 VC 200/35 LoopInd:0 
CTag:0x3F9B9 OAM Cell Type 5     


Is there any other queue where packets or cells may be queued for this long 
time?

Thank you
Bye,

[0] Routers are:
7200 + NPE300 + PA-A6-OC3 => STM-1 ATM
2800 + WIC Serial => Frame relay => ATM


-- 
  Daniele Orlandi



More information about the cisco-nsp mailing list