Re: TML Service Primitive Update

On Tue, 2006-07-11 at 20:46 +0800, Wang,Weiming wrote:

> How about others (also the authors) thoughts?
> 

I agree with Joel's view; in addition, the text still has a lot of
"howto" type of "features". 
I think you still have too many "features" that end up
tying things to some specific operations in some specific OSes. 
This document should be making recommendations, not say "this is how it
is done". It must never be imposing the how-to achieve things.
I sent you private emails on the topic already but didnt receive
responses - so a few of the thoughts are below:

1) In the events section where there are some events that are described
such as congestion alert, DOS alert etc. It is not defined how such
things get detected by the TML. It is also not mentioned how  such
information is used by the PL. An example of what would be useful would
have been something like:

"In cases where the OS is capable of detecting ECN (Early/explicit
congestion notification) where the underlying IP protocol is capable of
of passing information to the application, then the TML could pass such
information to the PL. The PL could use this information for example to
adjust its sending rates or increase or reduce the priority of certain
PL messages, etc"
 
This needs to be done for all events.

2) Again on events; you mention things like "sustained events". I think
the correct terminology is level-triggered vs edge-triggered. 
Again, there are a lot of OSes which may be capable of one vs the other.
You could say " It is suggested that error events are delivered as
level-triggered" but not impose they are delivered that way.

My recommendation is to really cut off a lot of this stuff off for now
and then we can have discussions where we can restore some pieces. 

There are a lot more issues like the above (which held me back from
editing before the deadline since i didnt hear from you). We also need
to improve readability in a big way.

cheers,
jamal