Hi Pekka,

 

I think this would work even in the case when the responses are
crossing.

Let each device send its response irrespective of whether it itself has
just sent the request.

 

Irrespective of who sent/received the response first, if we sync to the
higher values, then eventually both the devices will end up having same
values.

 

In the approach which I suggested, B need not wait for A to respond. It
can just send its values and ignore (don't update the received values )
the message response received from A

 

Please correct me if I am missing something. 

 

While arbitration can also be used, the above approach seems simple.

 

Regards,

Kalyani

 

 

-----Original Message-----
From: Pekka Riikonen [mailto:[email protected]] 
Sent: Thursday, September 16, 2010 3:14 PM
To: Kalyani Garigipati (kagarigi)
Cc: Yaron Sheffer; IPsecme WG
Subject: RE: [IPsec] Moving forward with the HA solution draft

 

On Thu, 16 Sep 2010, Kalyani Garigipati (kagarigi) wrote:

 

: Regarding the race condition, I think instead of starting the message

: Id's altogether with 1, we can have the devices sync up to the higher

: value among the responses.

: 

:   Cluster A              Cluster B

:     request SYNC      -->

:                 <--      request SYNC

:     response (4,4)    -->

:                 <--      response (5,5)

: 

: Cluster A should update its values to 5,5 whereas Cluster B should not

: update its values since its values are higher than the values at
Cluster

: A. 

: Which means Cluster B has correctly synced its message values  to some

: extent when compared with Cluster A.

: 

I don't think this can work, because the responses can cross each other
in 

the network also, and most likely do if the requests did as well.  That 

is, Cluster A responds same time as Cluster B before it has received the


response.  The implementation would have to wait for the response before


sending its own response.  But, what if the other end is waiting it
also?  

It's a dead lock.

 

I think it needs to be a pre-defined value that both end set in this
case 

or there needs to be some arbitration between the peers to deicde who 

waits.  I'd go for the simplest solution.

 

      Pekka

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

Reply via email to