Thanks for the response! Comments inline:
Thanks!
Ben.
On Jun 21, 2013, at 4:35 PM, Michael Thornburgh <mthornbu(_at_)adobe(_dot_)com>
wrote:
hi Ben. thanks for your review. comments/replies inline.
From: Ben Campbell [mailto:ben(_at_)nostrum(_dot_)com]
Sent: Thursday, June 20, 2013 4:07 PM
I am the assigned Gen-ART reviewer for this draft. For background on
Gen-ART, please see the FAQ at
<http://wiki.tools.ietf.org/area/gen/trac/wiki/GenArtfaq>.
Please resolve these comments along with any other Last Call comments
you may receive.
Document: draft-thornburgh-adobe-rtmfp-07
Reviewer: Ben Campbell
Review Date: 2013-06-20
IETF LC End Date: 2013-06-25
Summary: This draft is almost ready for publication as an informational RFC.
However, I have some
concerns about the purpose and intended status of the document that I think
should be considered prior
to publication.
Note: This is an informational draft that describes what I understand to be
an existing protocol as
implemented by commercial products. Therefore, this review does not attempt
to evaluate the protocol
itself. I assume the protocol is what it is, and it is not up to the IETF to
agree or disagree with
it.
*** Major issues:
-- Why does this need to be published as an IETF stream RFC? If I
understand correctly, this
documents an existing protocol as implemented by commercial products. I
agree with Martin's comment
that there is value in publishing this sort of thing, but I applaud the
Adobe and the author for
publishing it so other implementations can interoperate with their products.
But that could have done
that in an independent stream document, or even in an Adobe published
document. (Perhaps even in a
prettier format ;-) ) If we publish this as an IETF stream document, then
I think it needs stronger
clarification that it is not an IETF consensus doc than just its
informational status.
this memo documents an existing network transport protocol that is widely
deployed and in widespread use in the Internet. we felt that the IETF
community (and in particular the participants in the Transport and Services
Area) is the most appropriate community with which to share this work: 1)
members of this community have the necessary and relevant expertise not only
to understand the protocol, but to make use of it potentially in new
applications beyond Flash; 2) it is a transport protocol similar in many ways
to TCP and SCTP and widely deployed and used; 3) Adobe is interested in
pursuing standardization of this protocol (with all that entails) if there is
community interest, and the IETF is definitely the right place for that.
we are very grateful that Martin Stiemerling chose to sponsor this document.
with regard to comments in later messages in this thread, i'd be happy to
include some (IESG-supplied) boilerplate in the document to clarify that it
is not the product of an IETF WG. however, note that both the title and
first sentence of the Introduction indicate that this is "Adobe's
blahblahblah", consistent with other vendor-protocol RFCs and consistent with
IESG editorial insistence (as told to me by a former TSV AD). see RFC 4332
and RFC 6802 for two examples of vendor-specific/supplied protocols. see
also the IESG note in RFC 4332 as an example disclaimer that could be added.
Some additional text (whether IESG boilerplate or otherwise) that clarifies the
purpose of the draft would help a lot.
Along those lines:
-- Is this document the authoritative specification? (I suspect not.)
Who owns change control? I
assume that to be Adobe and/or the authors. What expectation do readers of
this draft have that it
represents the current version of RTMFP at any point in time?
this memo is the authoritative specification for Adobe's RTMFP. Adobe owns
change control. i believe the second and third paragraphs of the
Introduction indicate to a reader that this draft represents RTMFP as
deployed in the named Adobe products at the time of writing.
At the time of writing yes. My concern is how a future implementor can be
confident that this doc describes RTMFP as used by Adobe at points in the
future. When you say this is the authoritative specification, does that mean
that Adobe does not plan to modify the protocol without timely publication of
an update to this document?
-- Under what circumstances would one desire to implement this? I can
infer that from the
introduction, but it would be nice to see some sort of applicability
statement making it explicitly
clear that this is not an IETF protocol, and that one would implement it in
order to interoperate with
certain Adobe products. I know that some of this may be implied by the fact
that this is informational
rather than standards track. But I don't expect readers who are not IETF
insiders to understand that
subtlety without some explicit words to that effect. In particular, it would
be good to clarify the
difference between this and the many "not quite accepted as standards track
by some working group"
nature of a number of our informational RFCs that describe protocols.
i will expand on applicability beyond what i specify in the first paragraph
of the Introduction, in general terms such as the kinds of functionality we
use this protocol for in Flash. note that interoperation with certain Adobe
products, such as Flash Player, requires additional information such as the
Flash-specific Cryptography Profile and syntax/semantics of mapping Flash's
"Real Time Messaging Protocol (RTMP)" messages onto RTMFP flows. these
Flash-specific profiles and mappings are application-layer for RTMFP,
analogous to HTTP over TCP.
Okay
That all being said, this is overall a well written document. The rest of my
comments are mostly
pedantic nits.
*** Minor issues:
-- section 1.2:
I find the use of "MUST ONLY" confusing. I gather it means "you are
_allowed_ to do X only under
certain conditions" rather than "you are _required_ to do X under certain
conditions." If correct, I
think the words "MAY ONLY" would be more clear. If not, then I think the
construct would be better
handled using existing 2119 language.
you're not the first person to be confused by that construct. i will change
instances of "MUST ONLY" to "is allowed only if" (or similar) and remove the
definition for "MUST ONLY" from Section 1.2.
Works for me.
-- section 3.2:
Do I understand correctly that all endpoints are expected to be able to
present certificates? Do you
find that common in the field? I realize the nature of the certs will depend
on the security profiles.
yes, all endpoints have certificates. in the RTMFP-for-Flash world, these
certificates are not X.509 certs, and are anonymous and ephemeral.
Okay.
-- sections 3.2 and 5
Do I assume correctly that endpoints need a common crypto profile to
communicate? Is there a
repository where one might find crypto profile documentation? Is there a
commonly implemented one to
which this draft could refer?
yes, endpoints need a common cryptography profile to interoperate. there is
no repository for crypto profile documentation at this time. currently, there
is one defined cryptography profile (the one used for Flash communication)
that is not publicly documented, because i do not yet have permission to
publish it. i (meaning me personally) hope that a memo documenting the
crypto/application profile for Flash communication (as being a widely
deployed and used profile for RTMFP) would also be published someday as an
I-D and hopefully an Informational RFC.
I understand the issue about permission to publish, but does this document
serve its purpose until you are ready to publish the crypto profile? Ideally
they would be published together and this document would reference that one. Do
I infer correctly that there is no way someone could create an implementation
that interops with Adobe products based on the documents available at this time?
-- section 3.2: "Multiple endpoints SHOULD NOT have the same identity."
Why not MUST? Will things not break if two endpoints do have the same
identity?
this should be "It is RECOMMENDED that multiple endpoints not have the same
identity." if two endpoints have the same identity, then they will be
indistinguishable. this will not break RTMFP, but might confuse an
application. that being said, there could potentially be reasons to have
different endpoints with indistinguishable identities and that potential
should not be normatively prohibited.
As Barry mentioned, RECOMMENDED is a synonym for SHOULD. Adding some text the
effect of your additional explanation would solve my concern.
-- "Authenticity MAY comprise verifying an issuer signature chain in a
public key infrastructure"
Is that really normative, or just an observation of fact?
that "MAY" should be "can".
Okay.
-- " Canonical Endpoint Discriminators for distinct identities SHOULD be
distinct."
What if they are not? Do things break? It might be worth making this a MUST,
or adding some additional
guidance about what might happen if the SHOULD is violated.
i will add a note that if the canonical EPD is the same for two distinct
identities, then the "duplicate session" logic in section 3.5.1.1.1
(paragraph 6 step 1) might abort a new opening session to the second
identity, which might not be desired.
Okay.
-- "A comparison function MAY comprise performing a lexicographic
ordering..."
Is that really normative, or just an example of something one might do?
that "MAY" should be "can".
Okay.
-- "A test SHOULD comprise testing whether the certificates are identical."
What if it doesn't? Also, what constitutes "identical"? All fields match
values? Bitwise match? Is it
simply including the same identity or identities? Maybe an identity function
provided by the crypto
profile?
again, that SHOULD should be reworded to RECOMMENDED. i will change
"identical" to "bitwise identical".
See my earlier comment on SHOULD vs RECOMMENDED. The important thing is to
clarify the implications of violating the SHOULD.
-- 3.5, last paragraph:
Can you offer guidance on reasonable buffer size and number ranges?
i assume you mean "3.4, last paragraph" (referring to packet reassembly
buffers). i will add some guidance and a "for example". the guidance will
depend on how that RTMFP will be used and the expected amount of resources
available to it.
That would probably help. The important thing is to help implementers avoid bad
decisions on buffer size ranges.
-- 3.5.1.1.1, 3rd paragraph:
What are the consequences of having a tag with less than 8 bytes of length?
Is the SHOULD reasonable
here?
both of those SHOULDs should be RECOMMENDEDs.
See previous comment
-- 3.5.1.1.1 throughout, and following sections:
Does the upper case AND have special meaning?
upper case is the closest to bold face available in this format. the
intention is to emphasize (since there are multiple conditionals) that all
conditionals are required to hold.
Understood. I'm not sure if there is a convention for that--I guess the RFC
editor will either leave it or advise something else.
-- 3.5.1.4, Deployment Note:
How would a redirector know which endpoints might do this? Or should it
never initiate a session?
this is guidance for designing and deploying an RTMFP system. the redirector
wouldn't know -- the designer should design the system such that the
described circumstance doesn't happen (for example, by having the redirector
never initiate a session). this is properly a SHOULD NOT since doing so can
cause undesired behavior (as described). it is not desirable to give a
normative prohibition or recommendation against a redirector initiating any
sessions. i feel this paragraph gives the narrowest normative limitation and
an explanation for that limitation.
Okay. Some words to that effect would help.
-- 3.5.2:
Do I infer correctly that two endpoints need not share the same congestion
control algorithm to
communicate?
correct. the timestamp echo facility, ack behavior (as described for flow
receivers), and limitations on aggressiveness in 3.5.2/.* impose the
normative envelope in which any congestion control algorithms should
interoperate.
Okay.
-- 3.6.2.3.2, 2nd paragraph: "...an implementation SHOULD encode a Next User
Data chunk instead of a
User Data chunk"
I'm not sure why this needs to be normative, as I assume its just normal
behavior. But if it does need
to be normative, why not a MUST? Can the far endpoint reasonably handle
things if the SHOULD is
violated?
this should be a RECOMMENDED. the far end will work just fine if Next User
Data is never used.
See previous comments on SHOULD vs RECOMMENDED.
*** Nits/editorial comments:
-- General: There's quite a bit of inconsistent use of third-person vs
second-person language.
i will try to clean that up.
Okay.
-- 3.1: It would be nice to see the overview earlier in the draft. I found
it difficult to read
through all the data format stuff prior to section 3 putting them into
context.
others have commented that they'd prefer sections 2 and 3 to be swapped. i
resisted that as this order ("here are the pieces" and then "here's how they
fit together") is my preferred way of explaining stuff. also swapping those
sections would be a huge editorial change which might have significant
cross-reference and forward/back reference implications and could introduce
hard-to-detect errors into the spec. however, i think i can move the
overview (or a summary thereof) to earlier in the spec (probably in Section
1).
Okay.
-- 3.5.1.4, Deployment Note:
s/which/that
good catch.
-- 3.5.1.6, last paragraph:
Which diagram? (An explicit cross-ref would help.)
oops. that paragraph used to be the postamble of the figure, so it obviously
meant "this figure". i'll change "the figure" to "Figure 17".
Okay.
-- 3.5.2:
What is meant by "aggressive" in this context? Aggressive in avoiding
congestion, or aggressive in
sending data?
aggressive in sending data. i'll make that clarification.
Okay.
-- 3.6.2.3.1 and subsequent sections:
Does the all-caps "FIRST" have special meaning?
it is all-caps for typographical emphasis.
thank you!
-michael thornburgh