Re: Something improper with ASP state change procedures.

The systematic lost packets are probably caused by SCTP ending the data
between an IP address pair other than the one on which the connection
was established, and on which it hasn't successfully sent a heartbeat.
The SCTP stacks I've used seem to do this a lot, but to my mind they
should not, and the observered behaviour ought to be considered a bug.
 
At connection establishment time, SCTP seems to assume that the IP
addresses of one system should be able to send to all the addresses of
the other system, and that any required links are actually working.
This isn't a good assumption to make!
 
Consider the following 3 systems:
 
a) public IP 1.1.1.1, private network 192.168.1.1
 
b1) public IP 1.2.1.1, private network 192.168.1.2 (shared with machine
b2)
b2) public IP 1.2.1.2, private network 192.168.1.3 (shared with machine
b1)
 
An SCTP connection between 'b1' and 'b2' can use either network.
An SCTP connection between 'a' and 'b1' (or 'b2') must only use the
public addresses.
 
It isn't obvious to me how anyone is supposed to know the network
topology well enough to make this work properly.
 
    David


________________________________

	From: [email protected] [mailto:[email protected]]
On Behalf Of Naveen Kottapalli
	Sent: 23 November 2010 14:36
	To: [email protected]
	Subject: [Sigtran] Something improper with ASP state change
procedures.
	
	
	Hi All,
	 
	We are seeing some strange in the M3UA ASP state change
procedures.  Please find the description of the problem below.
	 
	ASP
SGP
	=======================================
	ASPUP               -------------->  Processes ASPUP and sends
ASPUP_ACK
	 
	Move ASP to
	INACTIVE state   <-------------   ASPUP_ACK
	 
	SCTP-SACK
	for ASPUP_ACK  ------------->  Missed or lost in the network
	 
	ASPAC               -------------->  Processes ASPAC and sends
ASPAC_ACK
	 
	Move ASP to
	ACTIVE state      <-------------   ASPAC_ACK
	 
	SCTP-SACK
	for ASPAC_ACK  ------------->  Missed or lost again in the
network
	 
	Move ASP to
	INACTIVE state   <-------------   SCTP layer at SGP retransmits
the ASPUP_ACK, ASPAC_ACK because of lost SACKs.
	because of
	ASPUP_ACK
	in ACTIVE state
	 
	SCTP-SACK        ------------->  Reached the SCTP layer of SGP
	 
	In the above behavior the M3UA at SGP thinks that the ASP is in
ACTIVE state whereas it is not the case.  Since ASP is in INACTIVE
state, all the data packets that were sent by SGP will be dropped
(without sending any error message back to network).  The above problem
will lead to a potential message loss in the network.
	 
	Considering the above behavior, I think the state machine should
be tuned accordingly.
	 
	Please pass on the comments on this.
	 
	Thanks & Regards,
	Naveen.

_______________________________________________
Sigtran mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sigtran