got some questions on support for multihoming

got some questions on support for multihoming

[Chris Friesen]

Hi all,

I've got a few questions about the current implementation of SCTP.
Suppose I have two network nodes, each of which has two IP addresses.
Theoretically this results in four possible paths between the two nodes.

If the network device corresponding to the primary IP address goes down
on the remote node, will the current implementation properly switch over
to use the "most divergent source-destination pair" as suggested in the
RFC?  I assume this would mean switching to the other IP address on both
nodes?

Same question, but with the device corresponding to the primary IP
address going down on the local node.

If a chunk times out, will the current implementation retransmit it via
the "most divergent source-destination pair" as suggested in the RFC?

Thanks,

Chris

Re: got some questions on support for multihoming

[Vlad Yasevich]

Chris Friesen wrote:
> Hi all,
> 
> I've got a few questions about the current implementation of SCTP.
> Suppose I have two network nodes, each of which has two IP addresses.
> Theoretically this results in four possible paths between the two nodes.

Theoretically, but often not in practice.

> 
> If the network device corresponding to the primary IP address goes down
> on the remote node, will the current implementation properly switch over
> to use the "most divergent source-destination pair" as suggested in the
> RFC?  I assume this would mean switching to the other IP address on both
> nodes?

It sort of depends on how the addresses are configured.  Assuming a typical
multihomed scenario where each address is in its own subnet, then the answer
is yes.  Ex:

node A (addrs: 1.2.3.4, 2.3.4.5)  <---------> Node B (addrs: 1.2.3.5, 2.3.4.6)

In this case, assuming 1.2.3.0/24 was primary, the path with switch to
2.3.4.5 <--> 2.3.4.5.

It really is a function of the routing looking by the kernel.

> 
> Same question, but with the device corresponding to the primary IP
> address going down on the local node.

Once the appropriate time-outs happen, it should switch the same as above.

> 
> If a chunk times out, will the current implementation retransmit it via
> the "most divergent source-destination pair" as suggested in the RFC?

Most of the time.  We simply try to pick a different destination address to
send to, and let the kernel pick a source.  As long as the source picked by
the kernel is within our bound set, we'll use it.  We don't check for
'mostly divergent pair'.

-vlad

> 
> Thanks,
> 
> Chris
> --
> To unsubscribe from this list: send the line "unsubscribe linux-sctp" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>