RE: [dhcwg] Order of occurence of DHCPv6 options

[Ralph Droms <[e-mail]>]

Unitl we have more experience with real implementations, it's hard to know 
exactly what to optimize.  So, let's run some experiments once we have 
interopable implementations and get some experience with real deployments 
so we can understand where the performance hits and improvement 
opportunities are.  Then we can write up that experience and a 
specification for option ordering...

- Ralph

At 03:49 PM 6/17/2002 +0530, Vijayabhaskar A K wrote:
>Hi Bernie,
>
>The requirement here is PERFORMANCE. Assume there are 1000 requests passing
>through a server which are NOT intended to be handled by this server. This
>will happen since most of the messages are Multicast only. This is not the
>corner case situation as you said, this will happen in most of the cases.
>There is nothing wrong in mandating this. This is something like specifying
>these options in a fixed mesg header. The server/client will ignore the
>remaining
>option field, if there is any error in the mesg header.
>
>All of the above, the relay should be lightweight, if the admin has
>configured
>the relay to load balance based on Elapsed time option and how bad the
>performance will be if the relay has to search for this option by parsing
>the whole message.
>
> > the above might not even apply (for example, if a client or
> > server always decodes all options to validate the packet,
> > it won't save anything).
>
>Then, this is a poor implementation. If the first option itself is
>erraneous, (for ex. auth itself fails), then there is no need to
>parse remaining stuff. We should leave these kind of implementation
>as corner case and go on mandating options' order of occurence.
>
>~ Vijay
>
>-----Original Message-----
>From: Bernie Volz (EUD) [mailto:[e-mail]]
>Sent: Saturday, June 15, 2002 11:06 PM
>To: '[e-mail]'; [e-mail]
>Subject: RE: [dhcwg] Order of occurence of DHCPv6 options
>
>
>Vijay:
>On the order of options, we chose specifically NOT to specify this. There
>really is no
>requirement or overwhelming reason to do this. Putting some options earlier
>than others
>improves performance in some situations so it will be wise for clients (and
>servers) to
>do this, but it is difficult to specify this and a client and server MUST
>assume the
>worst anyway.
>For example, I'd argue that putting the Server Identifier option first (or
>very early)
>is best since it allows all servers that receive this (except for the
>intended one) to
>drop the packet almost immediately. Rather than having to look through 4
>options, won't
>it be better if these servers only had to look at 1 option (the first).
>Though, having said that there is no reason to require this and depending on
>how a client
>or server processes packets, the above might not even apply (for example, if
>a client or
>server always decodes all options to validate the packet, it won't save
>anything).
>So, we specifically decided not to specify option ordering and I think that
>is a wise
>decision.
>
>
>Regarding the other issues, some clean-up for the draft will be required to
>remove and
>fix items based on the final status codes/option usage.
>- Bernie
>-----Original Message-----
>From: Vijayabhaskar A K [mailto:[e-mail]]
>Sent: Saturday, June 15, 2002 1:31 PM
>To: [e-mail]
>Subject: [dhcwg] Order of occurence of DHCPv6 options
>
>
>I think, we need to fix the order of options occuring in DHCPv6 messages
>for better processing of the messages.
>1. Elapsed Time option - Since based on this value only, the relay/server is
>going to decide the policy of reply.
>2. Status Code - In the case of Failure, this message can be directly
>ignored
>3. Authentication option - If authentication fails, nothing is needed to
>be processed.
>4. server-id - If the server id doesn't match, server can ignore it.
>5. client-id - If the client id doesn't match, client can ignore it.
>6. Rapid commit - Based on this, the server can determine whether it need
>to send Advertise or Reply.
>
>
>If any of these options are occuring, then they MUST occur in this order.
>----------------------------------------------------------------------------
>--
>ConfNoMatch and AddrUnavail are nowhere used. It can be removed.
>AuthFailed is also nowhere used. We need to add text saying if the
>server/client is not able to authenticate the other, then they
>should send reply with error code AuthFailed set.
>----------------------------------------------------------------------------
>--
>R-forw.    *                        *     *      *      *
>R-repl.    *                        *     *      *      *
>Why Authentication, User class, Vendor class, Vendor specific options
>are needed in Relay-fwd and Relay-reply message.
>~Vijay
>
>
>_______________________________________________
>dhcwg mailing list
>[e-mail]
>https://www1.ietf.org/mailman/listinfo/dhcwg
>
>
>_______________________________________________
>dhcwg mailing list
>[e-mail]
>https://www1.ietf.org/mailman/listinfo/dhcwg


_______________________________________________
dhcwg mailing list
[e-mail]
https://www1.ietf.org/mailman/listinfo/dhcwg