Re: [tram] Review of draft-ietf-tram-stun-pmtud-03

[Brandon Williams <[e-mail]>]

Hi Marc,

Thanks. It looks like most of my comments are resolved.

After re-reading, I've still got a few comments, at least one of which I 
see that I missed on the previous pass. My new comments are below. 
Nothing is in-line. Let me know if anything is unclear.

--Brandon

Section 4

I suggest moving the first paragraph "A client MUST NOT ..." to section
4.1.1, since it's about forming/sending a message. Seems like it would 
be a good last paragraph in that section.

Section 4.1.1 para 2

s/transctaion/transaction/

Section 4.1.1 para 3

My question about FINGERPRINT was meant to indicate that the text
should say why it's needed. I see you've added that explanation
elsewhere. Maybe add the same here too.

Section 4.2.1 para 1

s/The server/The client/

Section 4.2.5

I still find "the same as" in para 1 confusing here since it's not
applied to any header fields. Maybe "similar to"?

Also, I found para 2 a little difficult to read. How about:

   For each STUN Probe Indication or Request, the server retrieves the
   STUN FINGERPRINT value. For all other packets, the server calculates
   the checksum as described above. It puts these FINGERPRINT and
   checksum values in a chronologically ordered list that is sent back
   in the Report Response.

Section 5.2

I'm still uncertain about this section. The receiver behavior makes 
sense, but that's not really what I was questioning. It's not clear what 
you intend for the original probe sender whose support for PMTUD is 
being assumed by the receiver. How did that first machine know that it 
was safe to send its probe? This section seems to imply that it injects 
probes as a way to tell the receiver that it supports the protocol, 
though you're right, it doesn't actually say that so I think I just 
misunderstood your intention.

So, how does a machine that doesn't know whether the receiver supports 
the PMTUD protocol know whether it's safe to send a probe when it hasn't 
received one from the other party? Is the idea that one side or the 
other must receive this information via signaling or some other means?

I see the paragraph in section 4 that I previously missed which states 
the client MUST NOT send if it doesn't know the other party supports it. 
Maybe section 5.2 could say something like "In cases where only one 
endpoint (the initial client) knows that its peer endpoint (the initial 
server) supports PMTUD, the initial client sends a Probe Request or 
Indication to the initial server." That seems a little clunky, but I 
think you'll get the point of my suggestion.

Section 6.1 para 2

Maybe you could indicate that section 4.2.5 and 4.2.6 describe two
methods for acquiring and formatting the identifiers that could be used
for this purpose. It's clear to me that this is the case, but the
current text seems to imply that no options are provided by this
document.


On 02/16/2017 02:52 PM, Marc Petit-Huguenin wrote:
> Hi Brandon,
>
> Thanks again for the review, and sorry for the delay in responding.
>
> See my comments inline.  A new version will be published in a few minutes, please have a look too.
>
> Thanks.
>
> On 12/27/2016 11:27 AM, Brandon Williams wrote:
>> Hi all,
>>
>> Apologies for taking so long to complete a review of this one.
>>
>> Although my list of notes is long, there's nothing really substantive here. All of my comments are just suggestions to improve the document's clarity. Incorporate or discard as you see fit.
>>
>> Please let me know if any of the following is unclear.
>>
>> --Brandon
>>
>>
>> 1. Introduction
>>
>> I find the second paragraph a tiny bit difficult to read. Here's a
>> suggestion.
>>
>> OLD:
>>
>>   This document only describes how probing mechanisms are implemented
>>   with Session Traversal Utilities for NAT (STUN). The algorithm to
>>   find the Path MTU is described in [RFC4821].
>>
>> NEW:
>>
>>   This document describes probing mechanisms for Session Traversal
>>   Utilities for NAT (STUN) for use in implementing the Path MTU
>>   Discovery algorithm described in [RFC4821].
>
> Your text change a bit the meaning I was trying to convey in the original text.  What about something like the following text:
>
> New:
>
>    Not all UDP-based implement the Path
>    MTU discovery mechanism described in [RFC4821].  These protocols can
>    make use of the probing mechanisms described in this document instead
>    of designing their own adhoc extension.  These probing mechanisms are
>    implemented with Session Traversal Utilities for NAT (STUN), but
>    their usage is not limited to STUN-based protocols.
>
>>
>>
>> I also find the fourth paragraph a tiny bit difficult to read.
>>
>> OLD:
>>
>>   Additional network characteristics like the network path (using the
>>   STUN Traceroute mechanism described in
>>   [I-D.martinsen-tram-stuntrace]) and bandwidth availability (using the
>>   mechanism described in [I-D.martinsen-tram-turnbandwidthprobe]) can
>>   be discovered using complementary techniques.
>>
>> NEW:
>>
>>   Complimentary techniques can be used to discover addition network
>>   characteristics, such as the network path (using the STUN Traceroute
>>   mechanism described in [I-D.martinsen-tram-stuntrace]) and bandwidth
>>   availability (using the mechanism described in
>>   [I-D.martinsen-tram-turnbandwidthprobe]).
>
> Done.
>
>>
>> 2. Overview of Operations
>>
>> In paragraph three, the first "it" in the paragraph should be expanded.
>>
>> OLD: ... then it initiates Probe transactions, ...
>>
>> NEW: ... then the Client initiates Probe transacations, ...
>
> Done.
>
>>
>>
>> Same comment for the next paragraph.
>>
>> OLD: ... then it sends Probe Indications of various sizes ...
>>
>> NEW: ... then the Client sends Probe Indications of various sizes ...
>
> Done.
>
>>
>>
>> 4. Probing Mechanisms
>>
>> In paragraph six, I think the SHOULD requirement about retransmissions
>> belongs in section 4.1.1, which already describes this requirement in
>> other terms ("the Rc parameter ... is set to 3."). This part of section
>> 4 should stick to providing the MUST/MAY requirements for which of the
>> mechanisms are to be implemented.
>
> Sentence removed.
>
>>
>> Additionally, I think that paragraph six could be moved to be text
>> for the currently empty section 4.1. Simple Probing Mechanism. In
>> that case, paragraphs 7-10 could be moved to be text for the
>> currently empty section 4.2. Complete Probing Mechanism. I think
>> moving the high-level descriptions of the two mechanisms into their
>> respective sections improves the document flow.
>
> Done.
>
>>
>> Finally, in the last paragraph of the section, it states "... the
>> client adds a sequence number in front of each UDP packet sent."
>> I think you mean "... the client prepends the UDP data with a
>> header that provides a sequence number." At least, this is what the
>> description in the later section indicates to me.
>
> Done.
>
>>
>>
>> 4.1.1. Sending a Probe Request
>>
>> This section should indicate the STUN method type used for a Probe
>> Request (prefered) or forward reference section 8.1.
>
> Done, using your preference.
>
>>
>> Regarding authentication, why would an implementation authenticate
>> the Probe transaction? Is doing so recommended? If so, why? This is
>> discussed later in security considerations, but it was unclear to me at this point in the document.
>
> Rewrote the text with a bit more explanations.
>
>>
>> Regarding the FINGERPRINT attribute, why is it required?
>
> The STUN messages need to be sent together with the packets of the the other protocol, which can be anything, so the FINGERPRINT helps differentiate STUN packets.
>
>>
>>
>> 4.1.2. Receiving a Probe Request
>>
>> Here again, why is the FINGERPRINT attribute required?
>
> See above.
>
>>
>>
>> 4.1.3. Receiving a Probe Response
>>
>> OLD: If the Probe transactions times out ...
>>
>> NEW: If the Probe transaction times out ...
>
> Done.
>
>>
>>
>> 4.2.1. Sending the Probe Indications and Report Request
>>
>> Why MUST the Probe Indication be authenticated, when this is not
>> required (only allowed) for a Probe Request? Here again, it's discussed later, but was unclear to me here.
>
> Same than above, clarified the text.
>
>>
>> Also, same question as above about FINGERPRINT here.
>
> See above.
>
>>
>> This section should indicate the STUN method type used for a
>> Report Request (prefered) or forward reference section 8.1.
>>
>
> Done, using your preference.
>
>>
>> 4.2.3. Receiving a Probe Indication and Report Request
>>
>> I suggest that you forward reference the section that describes
>> construction of the IDENTIFIERS attribute in the last paragraph.
>
> Done.
>
>>
>>
>> 4.2.4. Receiving a Report Response
>>
>> The description of Probe Failure is a bit incomplete. It should
>> include a little more detail.
>>
>> OLD:
>>
>>   If the Probe Indication identifier cannot be found in the Report
>>   Response, this is interpreted as a Probe Failure, as defined in
>>   [RFC4821] Section 7.5.
>>
>> NEW:
>>
>>   If the Probe Indication identifier cannot be found in the Report
>>   Response but identifiers for other packets sent before or after
>>   the Probe Indication can all be found, this is interpreted as a
>>   Probe Failure as defined in [RFC4821] Section 7.5.
>
> Done.
>
>>
>>
>> 4.2.5. Using Checksums as Packet Identifiers
>>
>> The second paragraph states "The algorithm used to calculate the
>> checksum is the same as the algorithm used for the FINGERPRINT
>> attribute." I find this description a bit unclear, since the
>> description of the FINGERPRINT calculation in RFC5389 discusses
>> the STUN header fields, etc. Here, you're just talking about
>> generating a CRC-32 value XORed with 0x5354554e. It might be
>> more clear to just say that.
>
> Done.
>
>>
>>
>> 5.2. Implicit Probe Support Signaling Mechanism
>>
>> I'm a little concerned about the idea of just injecting probes
>> as a way to figure out whether the receiver supports them. It's
>> hard to know for every possible protocol that this will work and
>> will not break the protocol. Maybe add a cautionary note.
>
> No, the idea is not to inject probes to detect support.  Here we know that the receiver will support receiving probes, because it acted as a probe sender.  The sender of a probe implicitly signals that it can itself receive and process probes.
>
> Section 4 states that all implementations of this specification MUST implement the server side of both probing mechanisms, and this is how the implicit signaling mechanism can work.
>
>>
>>
>> 6.1. IDENTIFIERS
>>
>> Sections 4.2.5. and 4.2.6. each describe 32 bit identifiers. I
>> think section 6.1. should say a little more than just "contents
>> are opaque". I suggest that you describe specifically what the
>> attribute value will look like if one of the two 32 bit schemes
>> is used and declare that other protocols may generate different
>> values.
>>
>
> All right, that was the difficult part.  When trying to do what you suggested, I discovered that there was no mechanism to put the Probe packets in the list in the IDENTIFIERS attribute.  So the text in both section was rewritten to add that possibility.  The sections also contain the description of the content of the IDENTIFIERS attribute.
>
>>
>> 6.2. PMTUD-SUPPORTED
>>
>> For clarity, I suggest something a little more like the value
>> description of DONT-FRAGMENT from RFC5766.
>>
>> OLD: This attribute is empty.
>>
>> NEW:
>>
>>   This attribute has no value part and thus the attribute
>>   length field is 0.
>>
>
> Done.
>

-- 
Brandon Williams; Chief Architect
Cloud Networking; Akamai Technologies Inc.