RE: Artart telechat review of draft-ietf-alto-multi-cost-08

Hello Martin,

Thanks a lot for your review. The draft will be updated so as to address them.
Best regards,
Sabine

-----Original Message-----
From: Martin Thomson [mailto:martin(_dot_)thomson(_at_)gmail(_dot_)com]
Sent: 06 April 2017 04:08
To: art(_at_)ietf(_dot_)org
Cc: alto(_at_)ietf(_dot_)org; ietf(_at_)ietf(_dot_)org; 
draft-ietf-alto-multi-cost(_dot_)all(_at_)ietf(_dot_)org
Subject: Artart telechat review of draft-ietf-alto-multi-cost-08

Reviewer: Martin Thomson
Review result: Ready with Issues

Document: draft-ietf-alto-multi-cost-08
Date: 2017-04-06
Reviewer: Martin Thomson

This document describes how ALTO can be used to acquire cost maps with
multiple cost metric instead of a single metric.

The document very carefully deals with backwards compatibility with existing
ALTO servers and clients.  I don't anticipate many issues arising from the
deployment of this protocol.

I started reading -07, but finished with -08.  I checked that the issues I 
raise
still exist, but I'm not infallible.  Apologies if I get something wrong.

Minor issues: I've identified a few issues that are more than nits, these are
marked "IMPORTANT" below.

The abstract includes considerable justificiation.  An abstract only needs to
describe the *what*, not the *why*.  Thus, what is there could be simplified
considerably, e.g.,

  This document defines a new method for retrieving multiple cost metrics in
a
  single request for an ALTO filtered cost map.  It also defines improvements
  to the constraints that can be used for filtered cost maps.

Section 1

The introduction uses a bunch of odd terms.  Some of these are recognizable
from the ALTO specification, but some of the jargon seems unnecessary.  In
particular, "Internet View", "Provider Network region" and "Vector costs".
All of which I think that I understand, but they make the doc hard to follow.

Generally, I found the introduction quite hard to follow, both for that reason
and structurally.  The introduction could be a lot shorter and more
concise:

1. ALTO defines multiple cost types (and more are being defined).
2. Clients sometimes consume multiple cost types.
3. Requesting multiple cost types at the same time is more efficient (for
  several reasons).
4. This document defines how to do that.
5. Separately, when multiple cost types are present, more sophisticated
  filtering can improve efficiency further.
6. This document defines how to do that too.

Section 2

There are several items in the list here that are not used:
Application Client,
Network Service Provider, maybe more.  Please check and remove those
that don't apply.

The RFC 7285 section reference thing is unnecessary.

This document doesn't cite RFC 2119, but it uses the keywords.

Section 3.1

The example shows an empty cost-type, but the schema you define allows it
to be absent.  You REALLY need to pick one.  I don't believe that this is a
compatibility issue: once you have determined that a client supports multi-
cost, then you can do anything you like, just be clear about it.

Section 3.2

I found the argument about the ease of writing a parser to be quite
unconvincing.  However, a new media type that is largely the same as the
existing media type won't necessarily result in code duplication.

Just say what it is you expect to happen and don't try to be apologetic about
it.  What you have appears to be a workable design.

Section 3.5

This section is confusing.  You only need to say that you are not altering 
full
cost map resources in any way and that clients need to use filtered cost maps
if they want multiple costs at the same time.  (Obviously you could have, but
creating multiple resources with the full combinatorial mess caused by
combining many cost types is unwieldy.)  At a minimum, the second
paragraph here can be removed.

Section 3.6.2

IMPORTANT: You don't define what happens when a client provides "or-
constraints"
and "constraints" at the same time.  There are several valid options, but you
need to choose.

Section 3.6.3

It is probably worth explicitly noting that if "testable-cost-types"
does not
include values from "multi-cost-types", then those types can't be included in
"constraints"/"or-constraints".

Please explain the default value for the index for the "constraints"/"or-
constraints" express in this section in addition to where it is hidden in a 
note
later in the document.

Section 3.6.5

Uppercase for "must not" in the second paragraph.  (The "may" later in the
paragraph might be better as "can".)

In the example, the resource named "filtered-multicost-map" is provided for
legacy reasons only.  Why bother including "max-cost-types" and "cost-type-
names" at all when "filtered-cost-map-extended" includes all that and more?

Section 4.1.1

The definition of the schema here (and later) actually redefines the object
completely.  I found that confusing initially.  It would be good to identify 
the
*changes* from the base specification somehow.

Can testable-cost-type-names be present and empty if cost-constraints is
false?
The first part of the definition permits that, the second forbids it.

Section 4.1.2

The redefinition of PIDFilter is unnecessary.

IMPORTANT: pids is optional in RFC 7285.  Why the change?

I find the redefinition of the optionality of cost-type to be worthy of 
special
note.

In the definition of "or-constraints", you use a "database query"
where words
would suffice.

Section 4.1.3

I find the choice of value for "cost-type" to be problematic.  It is a string 
in the
base protocol, so changing to an empty object is likely to cause more issues
than simply omitting it.

Section 4.2 contains mismatched braces/parens for section references.

Section 4.2.2

IMPORTANT: This provides a definition for ReqFilteredCostMap that is very
different to that in the base specification.  I think that this should have 
been
ReqEndpointCostMap.

As before, repeating the definition of EndpointFilter is unnecessary.

Section 4.2.3 - see comment on 4.1.3

Section 5

I would be more comfortable if the examples used obviously-spurious
metrics (e.g., "cattle-head-count", "smell", "shoe-size", etc...) than these
metrics that are pretty plausible.  More so when you claim that they are
widely valued, which implies some sort of validity.

It should be relatively easy to populate Content-Length now that the
examples are "final".

Section 5.1

You have unmatched braces in "meta" due to the comment.

Section 5.2

Do you want to show one of the examples as having no cost values at all
across all the cost types?