Re: Strange LK-SCTP behavior in a multi-homing M3UA setup

Michael Tüxen <[email protected]> Mon, 21 Feb 2011 19:37:05 +0100
Newsgroups gmane.ietf.sigtran
Message-ID <[email protected]>
On Feb 21, 2011, at 5:29 PM, mosbah abdelkader wrote:

> Hi,
> 
> 
> I have the following setup:
>           - 1 M3UA ASP: with 2 IP addresses.
>           - 2 M3UA SGP: each one with 4 IP addresses.
> 
> 
> After launch, the ASP setups 4 associations: 2 with each SGP. Each association is multihomed with 2 IP addresses in both sides.
> 
> 
> The ASP sends traffic to the 2 SGP in a loadsharing configuration.
> 
> 
> Normally, after associating the ASP should align with the 2 SGP with the following procedure:
> 
>                SGP                             ASP1
> 
> 
>                 |                               |
> 
>                 |<-------------ASP Up-----------|
>                 |-----------ASP Up Ack--------->|
>                 |                               |
>                 |<------- ASP Active(RCn)-------|  RC: Routing Context
> 
> 
>                 |-----ASP Active Ack (RCn)----->|      (optional)
> 
>                 |                               |
>                 |-----NTFY(AS-ACTIVE)(RCn)----->|
>                 |                               |
> 
> 
> 
> 
> The problem is that LK-SCTP is sending SHUTDOWN without ANY REASON sometimes from ASP, sometimes from SGP1 and sometimes from SGP2. This behaviour disturbs the whole execution process.
> 
> 
> I have attached a wireshark capture for the ASP side.
> 
> 
> I am using OpenSUSE 11.1 and lksctp-tools-1.0.9-1.28.
> 
> 
> Can you please provide some help or hint to help me resolve this issue.
I guess you are using one to many style sockets (since LKSCTP would not bundle
DATA chunks with a COOKIE chunk on one to one style sockets) and the timer
to automatically close an association is configured to 4 seconds.

It also looks like the M3UA layer does not follow the state diagram in Figure 3
of RFC 4666. Frame 63 contains a COOKIE bundled with a ASPAC, which should not
be the case...

Best regards
Michael
> 
> 
> Thank you in advance.
> 
> --
> Best Regards
> 
> Abdelkader Mosbah.
> 
> <test-m3ua_21-02-2011_15-30.pcap>_______________________________________________
> Sigtran mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/sigtran