Re: [quicwg/base-drafts] QPACK improvement: wrap absolute index values (#1644)

[Mike Bishop <[e-mail]>]

Okay, that helps a lot.  I see where you're coming from.

My perspective was that MaxEntries is an absolute bound on the size of the table, so if entries are off the end of the table (the "Error area" in your diagram) the correct interpretation would be that they're in the future.  That always allows some room to insert without acknowledgement, because of the difference between real entries and the zero-size entries used to calculate MaxEntries.  I don't think it leads to confusion since you have to reference the table on a stream to have anything to get confused by 😉 and once you've referenced an entry you have a new pin in how far the table can advance without acknowledgement.

In any scheme, you can't reuse a slot until the *eviction* of the item which previously occupied it has been acknowledged.  If the window is 2*MaxEntries or more, that's at least a full turnover of the table and the limit on unacknowledged entries automatically implies that requirement.  If the window is MaxEntries, you have to explicitly track that dependency which will create the limit on how many unacknowledged entries can exist.  I think either is easy enough to reason about, but yours tracks a single value rather than a ring buffer of references, which is a decided advantage.

Okay -- I'm sufficiently convinced.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/1644#issuecomment-411920382