Comcast6.net PMTU broken

[Bill Owens]

On Tue, Apr 17, 2012 at 05:24:35PM -0400, Ryan Rawdon wrote:
> From a host behind a 6in4 tunnel a request to www.comcast6.net hangs.  It looks like a PMTU brokenness issue since the connection is established and the request is sent successfully.

If you have access to an ipfw-based machine (OSX definitely counts, other BSDs as well, but not Linux) then the definitive way to test is with scamper:

[cookiemonster:~] owens% sudo /usr/local/bin/scamper -F ipfw -I "tbit -M 1280 -u 'http://www.comcast6.g.comcast.net' 2001:558:1002:4:68:87:29:36"
tbit from 2620:f:1:1201:21b:63ff:fea4:4d92 to 2001:558:1002:4:68:87:29:36
 server-mss 1440, result: pmtud-success
 app: http, url: http://www.comcast6.g.comcast.net/
 [  0.050] TX SYN             64  seq = 0:0             
 [  0.073] RX SYN/ACK         64  seq = 0:1             
 [  0.100] TX                 60  seq = 1:1             
 [  0.150] TX                245  seq = 1:1(185)        
 [  0.174] RX                 60  seq = 1:186           
 [  0.174] RX               1500  seq = 1:186(1440)     
 [  0.174] RX               1500  seq = 1441:186(1440)  
 [  0.200] TX PTB           1280  mtu = 1280
 [  0.223] RX               1280  seq = 1:186(1220)     
 [  0.223] RX               1280  seq = 1221:186(1220)  
 [  0.250] TX                 60  seq = 186:1221        
 [  0.273] RX                500  seq = 2441:186(440)   
 [  0.273] RX               1280  seq = 2881:186(1220)  
 [  0.300] TX FIN             60  seq = 186:2441        
 [  0.323] RX                 60  seq = 4101:187        
 [  0.323] RX               1280  seq = 4101:187(1220)  
 [  0.323] RX                500  seq = 5321:187(440)   
 [  0.350] TX FIN             60  seq = 186:2881        
 [  0.374] RX               1280  seq = 5761:187(1220)  
 [  0.374] RX               1280  seq = 6981:187(1220)  
 [  0.399] TX FIN             60  seq = 186:4101        
 [  0.423] RX                500  seq = 8201:187(440)   
 [  0.450] TX                 60  seq = 187:5321        
 [  0.474] RX               1280  seq = 8641:187(1220)  
 [  0.474] RX               1280  seq = 9861:187(1220)  
 [  0.474] RX                500  seq = 11081:187(440)  
 [  0.500] TX                 60  seq = 187:5761        
 [  0.550] TX                 60  seq = 187:6981        
 [  0.573] RX               1280  seq = 11521:187(1220) 
 [  0.573] RX               1280  seq = 12741:187(1220) 
 [  0.574] RX                280  seq = 13961:187(220)  
 [  0.600] TX                 60  seq = 187:8201        
 [  0.623] RX                280  seq = 14181:187(220)  
 [  0.623] RX FIN            889  seq = 14401:187(829)  
 [  0.650] TX                 60  seq = 187:8641        
 [  0.700] TX                 60  seq = 187:9861        
 [  0.750] TX                 60  seq = 187:11081       
 [  0.800] TX                 60  seq = 187:11521       
 [  0.850] TX                 60  seq = 187:12741       
 [  0.900] TX                 60  seq = 187:13961       
 [  0.950] TX                 60  seq = 187:14181       
 [  1.000] TX                 60  seq = 187:14401       


Here's the odd thing though - the first time I tested, I saw the response above. Successive tests have resulted in 'pmtud-toosmall', which indicates that the server didn't send any individual packets above 1280 bytes. I wonder if there are more than one server behind a balancer, and somewhat different behavior?

Bill.