Implications of using smaller lollipop counter window

[nyrahul]

The draft observes issues with the lollipop counters and their use of persistent storage. Specifically, the draft observes that the linear part of the counter needs to be backed up in the persistent storage in order to avoid problematic conditions when the node is rebooted.

The linear part length is essentially controlled by the window length. The bigger the window, the more time a node spends in the linear part. This means more writes to the persistent storage.

It was discussed (in IETF106) that maybe we could reduce the window length from the default value of 16 to much lower.

Following things need to be analyzed:

1. Can the window for all the sequence counters be reduced alike? Is there any risk?
2. What does it mean in general to reduce the window? What are the implications?
3. Is it possible to do some experiments in Whitefield to understand the implications?

Another related subject is whether it is possible to identify if the node has rebooted based on counter values?