[j-nsp] iBGP convergence time

Josef Buchsteiner josefb at juniper.net
Tue Feb 6 02:32:39 EST 2007


        so  your  conclusion is that it is unrelated to private or
        public  ASN.  Can  you  look  if  there  are  a  lot  of
        retransmissions   during  the  convergence? Version 7.4R1
        was effected by an improper handling of RFC1323 which can
        result  in  very slow tcp throughput or even stalling the
        tcp session.

        Wether  you  are  effected  by this you can configure the
        following. You need to reset the TCP session to make sure
        non-tcp-rfc1323 is effective.

        josefb at minsk# show system internet-options
        no-tcp-rfc1323;
        no-tcp-rfc1323-paws;

        We have tracked this internal PR64682 and it is fixed in
        6.4R5 7.1R4 7.1R5 7.2R4 7.3R3 7.4R2 7.5R1 and higher.


        If  this  does  not  change  anything  I  would certainly
        suggest  to  look  first  at  the  sender and tcpdump the
        session,  look  for  drops  on the interface and so on. I
        would  be  very  interested to understand the symptom you
        see  and  get  to  an  explanation. ( if possible you can
        unicast me access details and we can check this together)

        thanks
        Josef


        
        
        

        


Tuesday, February 6, 2007, 12:08:16 AM, you wrote:

G>    
G>    
G> On Thu, 2007-02-01 at 09:07 +0100, Gniewko wrote:
 >> Hi All,
 >>
 >> I've already posted on JuniperForum.com but there was no answer so
 >> trying here (sorry for duplicate).
 >>
 >> There are 2 m7i, m7i_1 has a couple of EBGP sessions used for
G>  receiving
 >> full feed from the Internet. m7i_2 hovewer has only one IBGP session
 >> with m7i_1. Configuration of IBGP looks like this (no special things):
 >>
G>  [...]
G>  
G>  Hello again,
G>  
G>  After couple tests (private vs. public ASN convergence time) conclusion
G>  was quite simple (what does not mean obvious/clear for me):
G>  private - fast
G>  public - slow
G>  The latest tests hovewer, bring the oposite deduction (private - slow,
G>  public - fast) what makes me quite sure that the problem has nothing in
G>  common with the ASN nature (perhaps it's good cause doesn't sound so
G>  spooky).
G>  Anyway, there is something wrong with the mentioned iBGP session.
G>  
G>  --
G>  Regards,
G>  G.
G>  
G>  
G>  
G>  _______________________________________________
G>  juniper-nsp mailing list juniper-nsp at puck.nether.net
G>  https://puck.nether.net/mailman/listinfo/juniper-nsp
G>   
G>   
G>    
 


More information about the juniper-nsp mailing list