Discourage use of flow control

[martinthomson]

Rather than specify a flow control algorithm, we should simply discourage its use.

Mark, I think that we resolved this one at the interim. I'd just like to confirm with you before proceeding.

[mnot]

We discussed giving advice about the use of flow control in a separate "implementation advice" document. From my notes:

--------->8---------

Appropriate Use of Flow Control

Flow control is defined to protect deployments (whether it's server, client or intermediary) that are operating under constraints. For example, a proxy must share memory between many connections. Generally, it addresses cases where the receiver cannot process input (for whatever reason), yet wants to be able to process other streams.

A valid strategy is to turn flow control off if you don't need these properties.

Strategies for Flow Control

1. In most cases, it's recommended to advertise an infinite window.
2. If you must, you can protect resources (e.g., memory) in your implementation. However, if this is done without knowledge of BDP, this may lead to non-optimal use of bandwidth.
3. When flow control is used with knowledge of BDP, it can both protect resources and assure utilisation of the connection is unimpeded. Note that this can be difficult to implement.

-----8<-------

Since we don't have a separate document yet, I'm OK with putting this in the spec for the time being; it might be helpful to mark it somehow (even just with a comment in source) that it's potentially splittable to a separate doc.

Flipping to editorial.

[martinthomson]

We still don't know how to set an infinite window...

[martinthomson]

Problem solved. d8013a3