Re: IPFIX Information Element in the ALTO protocol

On Wed, Jun 13, 2012 at 5:36 AM, Brian Trammell <trammell <at> tik.ee.ethz.ch> wrote:
> Hi, Rich, all,
>
> On Jun 13, 2012, at 2:21 PM, Richard Alimi wrote:
>
>> On Tue, May 15, 2012 at 10:10 AM, Brian Trammell
>> <trammell <at> tik.ee.ethz.ch> wrote:
>>> Hi, Rich, all,
>>>
>>> The additions you're talking about seem on a first pass like they'd fit in the IPFIX information model.
With respect to the definition of description of an Information Element, the emphasis on "flow" is really
a historical thing. Basically, description should be specified to be human-readable and detailed
enough that a reasonably expert person would know how to generate and/or interpret it. We'll remove
references to flow from the definition of description in RFC 5102-bis.
>>>
>>
>> The text regarding it being information being available to the
>> observer would be too limiting for the ALTO case too.
>
> Hm... how can a device export information about something it can't observe / doesn't know?

ALTO may know information about the endpoint, but not observe data
packets to/from it.

(I interpreted 'observer' to be the entity that contains an
Observation Point as defined in RFC5101.)

>
>> If both of those requirements were removed, what are the guidelines on
>> new information elements that would be accepted into the information
>> element registry?
>
> http://tools.ietf.org/html/draft-ietf-ipfix-ie-doctors-03 mostly covers this, though it handles
the case of applying the entire IPFIX protocol (see section 3) as opposed to just using the information
model for a new application/protocol. Section 4 contains detailed guidelines for defining new
Information Elements (summarized in Section 7 in checklist form), though again this is geared toward IEs
for usage with the IPFIX protocol.
>
>>> There's a more subtle issue with potentially applying the IPFIX information model to ALTO: the
possible need to specify an alternate encoding for IPFIX IEs represented in ALTO. Briefly: each IPFIX
Information Element has a type (these cover address and time types as well as the basic primitives,
strings, and raw octet arrays), and these specify encodings appropriate for using these types in IPFIX,
as well: integers in big-endian binary, times as big-endian binary integers since the epoch or in NTP (era
0) format, and so on. As ALTO is JSON-based, these representations are somewhat less than appropriate.
>>>
>>> Indeed, this could probably be addressed with a single specification of representations of IPFIX data
types in ALTO (or indeed, for textual protocols in general). This would probably not have any surprises
(v4 addresses in dotted-quad, v6 addresses as per RFC 4291, timestamps as per RFC 3339, and so on); it would
just need to be written.
>>>
>>
>> Thank you very much for the proposal for using this registry in ALTO.
>> My personal opinion is that using IPFIX information element registry
>> doesn't seem to be a great match to replace ALTO's endpoint property
>> registry.  The win doesn't seem to be that great to me, and it also
>> requires writing a new spec to map the datatypes.
>
> Indeed; do let me know if I can be of any further assistance. For what it's worth, I suspect we'll define
text-friendly representations for the IPFIX datatypes sooner or later anyway.
>
> Cheers,
>
> Brian
>
>> If others in the ALTO WG feel differently, then please speak up.
>>
>> Thanks,
>> Rich
>>
>>> Best regards,
>>>
>>> Brian
>>>
>>> On May 14, 2012, at 12:36 PM, Richard Alimi wrote:
>>>
>>>> On Mon, May 14, 2012 at 6:34 AM, Benoit Claise <bclaise <at> cisco.com> wrote:
>>>>> Hi Richard,
>>>>>
>>>>> On Thu, May 10, 2012 at 10:37 AM, Benoit Claise <bclaise <at> cisco.com> wrote:
>>>>>
>>>>> Dear all,
>>>>>
>>>>> I've been reviewing draft-ietf-alto-reqs-14 and the following requirement
>>>>> got me thinking:
>>>>>
>>>>>  REQ.  ARv14-5: An ALTO client protocol MUST be extensible to enable
>>>>>   support of other host group descriptor types in future.  An ALTO
>>>>>   client protocol specification MUST define an appropriate procedure
>>>>>   for adding new host group descriptor types, e.g., by establishing an
>>>>>   IANA registry.
>>>>>
>>>>> Why don't you reuse an existing registry, in which you will have all the
>>>>> Information Elements already defined instead of defining a new one?
>>>>> WhatI have in mind: the IPFIX I.E. IANA registry.
>>>>> The piece of information I see in draft-ietf-alto-reqs-14 are IP address,
>>>>> prefix, BGP AS: they're in IANA.
>>>>> And many other IEs are already present, for future ALTO extensions, if
>>>>> required.
>>>>>
>>>>> This would not only save one registry (ALTO Endpoint Property in
>>>>> draft-ietf-alto-protocol-10), but offers a bigger advantage.
>>>>> Let me explain.... When you will control your applications with ALTO, you
>>>>> will anyway want to apply a flow measurement to monitor your changes, and to
>>>>> serve as a feedback loop for more optimizations.
>>>>> And the chances are high that you will using a NetFlow/IPFIX based
>>>>> mechanism. Both NetFlow and IPFIX use the IPFIX I.E. IANA registry.
>>>>>
>>>>> Therefore, it would make sense to have consistent data models between ALTO
>>>>> and IPFIX, and avoid a data model proxy if we want to compare the data.
>>>>> Proposal: reuse the ElementIDs found in the IPFIX I.E. IANA registry
>>>>> somewhere in your protocol.
>>>>>
>>>>> The suggested use case is in place of the Endpoint Property registry?
>>>>>
>>>>> I'm referring to the examples in
>>>>> http://tools.ietf.org/html/draft-ietf-alto-protocol-11#section-7.5.3
>>>>> So I guess it's
>>>>> http://tools.ietf.org/html/draft-ietf-alto-protocol-11#section-10.3.
>>>>> However, http://tools.ietf.org/html/draft-ietf-alto-protocol-11#section-10.3
>>>>> doesn't contain the initial series of values.
>>>>
>>>> Those are two different things in the protocol.  The former is for
>>>> identifying endpoints themselves, the latter is for properties
>>>> associated with those endpoints.
>>>>
>>>>>
>>>>>
>>>>> Just to be sure I understand what would happen if we went that route,
>>>>> what are the restrictions for adding new information elements?
>>>>> Specifically, what kinds of elements are allowed or disallowed?
>>>>>
>>>>> Things that have been informally discussed in the ALTO WG so far have
>>>>> been things like load information and available network capacity.  Do
>>>>> those fit within the ipfix model?
>>>>>
>>>>> The IANA considerations for new IPFIX IEs is in RFC5102,
>>>>> http://tools.ietf.org/html/rfc5102#section-7.1.
>>>>> It mentions: IANA through Expert Review [RFC2434]
>>>>> The most up to date document regarding IE allocation is
>>>>> http://tools.ietf.org/html/draft-ietf-ipfix-ie-doctors-02.
>>>>> I don't believe there are any limitations on for new IEs such as load
>>>>> information and available network capacity.
>>>>> Copying the IPFIX chairs and Brian Trammell.
>>>>
>>>> I think the main consideration here is whether IPFIX really is the
>>>> canonical source (and is planned to be in the future) for data
>>>> that an application (note: not only network operator) would want to
>>>> know about an endpoint.  In particular, in the CDN use cases, an ALTO
>>>> map may be provided by a CDN to advertise costs to/from and properties
>>>> of various cache nodes.
>>>>
>>>>> From RFC5102, Section 2.1 I see that a registration seems to enforce
>>>> that the Information Element is derived from a flow or someone
>>>> observing the flow on the network:
>>>>
>>>>   description - The semantics of this Information Element.  Describes
>>>>      how this Information Element is derived from the Flow or other
>>>>      information available to the observer.
>>>>
>>>> To me, that seems too restrictive for the ALTO model which is
>>>> providing information to the application layer.
>>>>
>>>> Thanks,
>>>> Rich
>>>>
>>>>>
>>>>> Regards, Benoit.
>>>>>
>>>>>
>>>>> Thanks,
>>>>> Rich
>>>>>
>>>>> Disclaimer: I have not read the protocol details
>>>>>
>>>>> Regards, Benoit.
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> alto mailing list
>>>>> alto <at> ietf.org
>>>>> https://www.ietf.org/mailman/listinfo/alto
>>>>>
>>>>>
>>>>>
>>>
>