[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

comments on draft-ietf-psamp-sample-tech-04.txt

Dear all,

Here are some comments.

1. Will this draft be an information RFC or standard track?
If standard track, we need a normative and an informative reference section

2. Must have an introduction paragraph regarding the other PSAMP documents.
See also the message https://psg.com/lists/psamp/psamp.2004/msg00024.html, point 7 of the non major issue
As pointed out in the message https://psg.com/lists/psamp/psamp.2004/msg00024.html, point 3.
The reference would be more readable if there were [PSAMP-PROTO], [PSAMP-INFO], [PSAMP-FRAME], etc...

3. Terminology section.
The following paragraph should also be copied over in [PSAMP-FRAME], expressing
- that all terms are the same
- that all terms might not be used 

     The PSAMP terminology resulted from joint discussions of the
     authors of this document together with the authors of [Du04].
     Therefore all terms used throughout this document represent the
     common understanding of the authors of both documents and are
     consistent with those defined in [Du04]. Furthermore, it is
     aimed at consistency with the terms used in [QuZC03].

Regarding the terminology section, see also the points major 5, 6, 7, 8 and non-major 2, 4 of the message https://psg.com/lists/psamp/psamp.2004/msg00024.html

4. Section 3 about "Scope and Deployment of Packet Selection Techinque"
"Sampling" is in section 3.1
but "Filtering! is in section 4 and not in 3.2
Now, I agree the Filtering section is a big one!
    - Sampling in section 3
    - Filtering in section 4
    - Remove the intro couple of paragraph from section 3, which are anyway repeated in the sampling and filtering sections

The only problem with this new structure is that the following paragraph must find a new home.

     Packet selectors can be part of an IPFIX metering process and 
     can also use the IPFIX exporting process. This is expressed as 
     association to one or more IPFIX processes. 

I think the table in section 3 is confusing with the last column category because it contains too much information.
Why not have 2 tables. One for Sampling and one for filtering in the respective sections (see point 4)?
The initial proposal was:

                                     |  content-independent  |  content-dependent
      Sampling Scheme                |       sampling        |      sampling  
      systematic sampling:           |                       |  
      count-based                    |           X           |  
      systematic sampling:           |                       |    
      time-based                     |           X           | 
      random sampling:               |                       |   
      n-out-of-N                     |           X           |  
      random, probabilistic sampling: |                       |  
      uniform probabilistic          |           X           |    
      random, probabilistic sampling: |                       |  
      non-uniform probabilistic      |                       |          X 
      random, probabilistic sampling: |                       |  
      non-uniform flow-state         |                       |          X 

And just add a last column for deterministic selection, as you have in your draft.
Something similar for the Filtering table

5. Section 4.1 "mask/match filtering"
In practice, this means most of the time configuring an access-list if we speak about the packet header information.
I think it should be clearly mentioned here, even if this is briefly mentioned in the section 4.3 "Router State Filtering"
Same remark for the section 5.2 "information model for filtering techniques", we want to mention the "access-list ID" for "case: Matching"

6. Collector must be in the terminology section.

7. Section "Consistent Packet Selection" 
See point 10 non-major of https://psg.com/lists/psamp/psamp.2004/msg00024.html.
This is a duplicate with the [PSAMP-FRAME] draft and the right place is in this draft.

8. Section 5.1 "information model for sampling techniques"
     Case Systematic Time Based: 
        - Interval length (in usec), Spacing (in usec) 
I think the spacing unit should be in number of packets.
How to specify: I want to specify 1 packet every x msec.
Spacing in usec depends too much on the interface bandwidth

9. In section 6.2 "stratified sampling", an example would ease the understanding

10. A list of editorial changes that I will explain live to Tanja while in Seoul.

Thanks Tanja for the new version of this draft.

Regards, Benoit.

30.  See if my discussion with Maurizio is in there.

     Note that a common technique to select packets is to compute a
     Hash Function on some bits of the packet header and/or content
     and to select it if the Hash Value falls in  the Hash Selection
     Range. Since hashing is a deterministic operation on the packet
     content, it is a filtering technique according to our
     categorization. Nevertheless, hash functions are sometimes used
     to approximate random sampling. Depending on the chosen input
     bits, the Hash Function and the Hash Selection Range, this
     technique can be used to approximate the random selection of
     packets with a given probability p. It is also a powerful
     technique to consistently select the same packet subset at
     multiple observation points [DuGr00]