[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Design decisions made at the interim SHIM6 WG meeting



Iljitsch van Beijnum wrote:
On 20-okt-2005, at 6:54, Erik Nordmark wrote:

8.  Do not use locator ordering and index references in SHIM6 control
    messages in the initial base spec


I'm not sure we really decided that at the Interim meeting. In the current proto draft the locator preferences are expressed by assuming that the locator list is ordered so that the relative position can be used in the locator preferences without having to include all the locators themselves in order to express the preferences.


The advantage of doing it this way is that some messages will be shorter because they only need a few bits as a pointer to a locator rather than the full 16 bytes locator itself.

However, the disadvantage is that you must always be certain that the locator lists are in sync. In the presence of lost and reordered packets this means a significant extra burden on implementations.

I don't think the savings in relatively sparse signalling messages is worth the extra complexity.

I made the point in a follow-up mail that you need to worry about the complexity in any case.

Here is a contrived example:

A and B setup the context with locators A1, A2 and B1, B2. At least A is using CGA.

A gets a new locator (A3) and sends an update message. It doesn't hear an ack. Retransmits. Still nothing.

A then sees that the A3 locator has locally failed (e.g., flaky WiFi interface), so it wants to send a locator preference update saying "don't use A3".

But it hasn't yet seen an ack to the update that introduced A3.

Somehow the protocol needs to deal with this dependency, which adds some complexity.

There are two rather different approaches to solve this dependency between the multiple updates:
 - use sequence numbers and a window (=1?) for update messages, so that
   the first update will be retransmitted until acked, before the second
   update can be sent.

 - Putting the newest information in a single update message and start
   sending that one. Thus including both a new locator list (which
   includes A1, A2, A3) and a preference option (which says "don't use
   A3"). This is then retransmitted until ACKed.

  Erik