I am the assigned Gen-ART reviewer for this draft. The General Area
Review Team (Gen-ART) reviews all IETF documents being processed
by the IESG for the IETF Chair. Please wait for direction from your
document shepherd or AD before posting a new version of the draft.

For more information, please see the FAQ at

<

http://wiki.tools.ietf.org/area/gen/trac/wiki/GenArtfaq>.

Document: draft-ietf-i2rs-traceability-09.txt
Reviewer: Elwyn Davies
Review Date: 2016/05/05
IETF LC End Date: 2016/04/29
IESG Telechat date: 2016/05/05

Summary:


I have concerns about the trace model used as explained below.  It may 


be that there is good reason and WG consensus for the model adopted, but 


it would be good to see some explanation of the rather curious hybrid 


model used.  There are also significant issues with the description of 


timestamps and a number of other nits/editorial matters to address.




Apologies for the last minute delivery.

Possibly Major issues:


Trace model:  The tracing model seems to be a curious hybrid of state 


recording and event logging.  The introduction seems to imply that the 


tracing model records events.  Indeed it does but state entry events do 


not appear to get recorded until the sequence transitions out of the 


state.  I can see that the COMPLETED entries record the total processing 


period, but this loses the detail of when actual processing of the event 


starts (as opposed to becoming PENDING).  I was somewhat surprised that 


a simple chained transition event model was not used (especially since 


the tracing entries are actually chained together already).






In particular if some sort of disaster occurs, it seems possible in this 


model that events in the PENDING queue might never appear in the trace 


log at all if the request hasn't started being processed. It also 


doesn't record any preprocessing time before the request becomes 


PENDING.  If there is a processing bottleneck this could be significant 


information.






I was also wondering whether this model traces the arrival and departure 


of clients (and whether authoentication/authorisation worked or not).   


This may be covered by operation types in the architecture which I 


haven't had time to read in detail.




Minor issues:

Nits/editorial comments:


s1:  The Intro should also contain a description of the intention of the 


document - basically a slight reworking of the abstract.  It should also 


outline the association of the framework with the interface (i2rs 


client<->agent) to which the traceability applies.




s3:



The
    ability to automate and abstract even complex policy-based controls
    highlights the need for an equally scalable traceability function to
    provide event-level granularity of the routing system compliant with
    the requirements of I2RS (Section 5 of
    [I-D.ietf-i2rs-problem-statement]).



The 'routing system' doesn't have an event-level granularity.  Maybe
OLD:
provide event-level granularity of the routing system
NEW:


provide recording at event-level granularity of the evolution of the 


routing system



END

s4:  The section ends with this list of 'use cases':



    As I2RS becomes increasingly pervasive in routing environments, a
    traceability model offers significant advantages and facilitates the
    following use cases:

    1  Automated event correlation, trend analysis, and anomaly
       detection;

    2  Trace log storage for offline (manual or tools) analysis;

    3  Improved accounting of routing system operations;

    4  Standardized structured data format for writing common tools;

    5  Common reference for automated testing and incident reporting;

    6  Real-time monitoring and troubleshooting;

    7  Enhanced network audit, management and forensic analysis
       capabilities.



I have added numbers to facilitate these comments:


IMO #2 and #4 are either not use cases or a not phrased as use cases.  


The automated testing is not really a use case as such. Having these 


characteristics supports the implementation of the actual use cases.  


Related to the data retention comment above, storing some or all of the 


trace log - and knowing which bits might be critical to control data 


retention - is a use case but the basic storage is just a necessary 


prerequisite of doing other things.  I also might suggest a reordering 


indicating importance perhaps.




Thus I would suggest replacing this with something like:

   As I2RS becomes increasingly pervasive in routing environments, a
   traceability model that supports controllable trace log retention


   using a standardized structured data format offers significant 


advantages,


   such as the ability to create common tools and support automated 


testing,



   and facilitates the following use cases:

   o  Real-time monitoring and troubleshooting of router events;

   o  Automated event correlation, trend analysis, and anomaly
      detection;

   o  Offline (manual or tools-based) analysis of router state evolution
       from the retained trace logs;

   o  Enhanced network audit, management and forensic analysis
       capabilities;

   o  Improved accounting of routing system operations; and



   o Providing a standardized format for incident reporting and test 


logging.






s5: .. is empty: Empty sections are not desirable.  A brief overview of 


the following sub-sections should be added (or alternatively promote 


s5.1 which actually describes the framework).




s5.1, para 1:



Some notable elements of the architecture are in
    this section.


I don't understand this sentence.  If it implies that elements of the 


architecture are defined in this section then one has to ask 'Why aren't 


they defined in the architecture document?'  Since s5.1 contains the 


whole framework, what other elements than the 'some notable' ones are there?






s5.1, para 2: The term 'northbpund' is not defined (and isn't used in 


the architecture').






s5.2: The title is ' I2RS Trace Log Mandatory Fields'  - nothing that 


isn't mandatory is discussed.  Should there be some words about optional 


extra fields?






s5.2, timestamps:  The RFC3339 format doesn't tie up with 32 bit 


resolution - there are hours and minutes etc and decimal representation 


is used.  Things like origin for timestamps needs to be defined if they 


are to be truly useful for comparison outside an individual enterprise 


(as might be implied by the incident reporting use case).  If RFC 3339 


format is really used, then the timestamps need to include the date as 


well since logs will certainly run over more than one day.  I note that 


the example in s6 shows full RFC 3339 date/time format examples.






s5.2, Applied Operation Data:  Does the Operation Data Present flag 


apply to this field?  Can this be present even if there is no Requested 


Operation Data?




 s5.2, Result Code: Need to expand acronym RIB.



s7.2:  One key point about timestamping (motivated by bitter experience) 


is that timestamps need to be recorded at the point when the event 


actually happens and not when the event is (potentially significantly 


later) entered into the log.  Logging is (as indicated) often allocated 


a low priority and event log writing may end up being postponed for a 


considerable time.






s11: I would consider I-D.ietf-i2rs-problem-statement and 


I-D.ietf-i2rs-pub-sub-requirements to be Informative; and


I-D.ietf-i2rs-rib-info-model, RFC 3339 and possibly RFC 5424 to be 


normative.