Re: [mpls] draft-ietf-mpls-mna-hdr ready to start the WGLC process

["jmh.direct" <[e-mail]>]

Thanks.  Sounds good to me.Yours,JoelSent via the Samsung Galaxy S20 FE 5G, an AT&T 5G smartphone
-------- Original message --------From: Tony Li <[e-mail]> Date: 8/4/23  7:31 PM  (GMT-05:00) To: "Joel M. Halpern" <[e-mail]> Cc: Haoyu Song <[e-mail]>, [e-mail] Subject: Re: [mpls] draft-ietf-mpls-mna-hdr ready to start the WGLC process Side nit: The acronym MSD is already taken by “Maximum SID Depth”, which has slightly different semantics.We already have another appropriate term defined, Entropy Readable Label Depth (ERLD) which actually has the semantics that we’re after, tho of course, we are not concerned just with Entropy.I would like to propose that we define Readable Label Depth (RLD) to be a synonym for ERLD, retroactively, and use that term hereafter.TonyOn Aug 4, 2023, at 2:26 PM, Joel Halpern <[e-mail]> wrote:
  
    
  
  The working group has so far been clear that maximum visible
      stack depth (MSD) matters.  If the WG changes that view, then
      various things are possible.  As long as we take that view, having
      multiple occurrences of ISD is necessary.  
    Yours,Joel
    
    On 8/4/2023 5:17 PM, Haoyu Song wrote:
    
    
      
      
      @font-face
	{font-family:SimSun;
	panose-1:2 1 6 0 3 1 1 1 1 1;}@font-face
	{font-family:"Cambria Math";
	panose-1:2 4 5 3 5 4 6 3 2 4;}@font-face
	{font-family:Calibri;
	panose-1:2 15 5 2 2 2 4 3 2 4;}@font-face
	{font-family:"\@SimSun";
	panose-1:2 1 6 0 3 1 1 1 1 1;}p.MsoNormal, li.MsoNormal, div.MsoNormal
	{margin:0in;
	font-size:11.0pt;
	font-family:"Calibri",sans-serif;}a:link, span.MsoHyperlink
	{mso-style-priority:99;
	color:blue;
	text-decoration:underline;}span.EmailStyle18
	{mso-style-type:personal-reply;
	font-family:"Calibri",sans-serif;
	color:windowtext;}.MsoChpDefault
	{mso-style-type:export-only;
	font-size:10.0pt;
	font-family:"Calibri",sans-serif;
	mso-ligatures:none;}div.WordSection1
	{page:WordSection1;}
      Hi Greg,Thanks for the explanation, although I
          think my point is perfectly technical and specific.We are designing a protocol for dataplane.
          Therefore I think the simplicity and correctness
          considerations are critical. The procedure described by Joel
          does bind to an implementation and deployment scenario with
          the implication of multiple NAS copies in the stack. If we
          assume to support only a single NAS, the procedure could be
          quite different, which also needs to be documented. If we
          reject such an idea, we need specific technical reasons.    Best,Haoyu 
        From: Greg Mirsky
            <[e-mail]> 
            Sent: Friday, August 4, 2023 1:54 PM
            To: Haoyu Song <[e-mail]>
            Cc: Tony Li <[e-mail]>; jmh.direct
            <[e-mail]>; Joel M. Halpern
            <[e-mail]>; Stewart Bryant
            <[e-mail]>; [e-mail]
            Subject: Re: [mpls] draft-ietf-mpls-mna-hdr ready to
            start the WGLC process
         
        
          Of course.
            You've said:
            
            
              
                My
                      gut feeling is that maintaining a single NAS
                      requires fewer lines of dataplane processing code.
                      It also keeps the header overhead minimum which
                      can potentially avoid other troubles such as MTU
                      limitation. So I suggest we actually do such
                      evaluations before deciding the final scheme.
                
              You, as a
                  seasoned developer, may decide to allow only a single
                  NAS in the stack, while other implementors may decide
                  differently. But, in my opinion, the architecture and
                  the solution MUST NOT preclude any of such decisions
                  (and it doesn't, AFAICS). And consideration of the
                  Path MTU before pushing a new NAS onto a packet is, in
                  my opinion, a common practice (at least it should be),
                  not specific to MNA. Hence my invitation to you,
                  please separate implementation and deployment issues
                  from what really impacts the MNA architecture and
                  solution.
            
            I hope
                  that my position is clearer to you now.
            
             
            
            Regards,
            
            Greg
            
          
         
        
          On Fri, Aug 4, 2023 at 1:42 PM Haoyu
              Song <[e-mail]>
              wrote:
          
          
            
              
                Greg, Please
                    be specific which of my comment is not technical and
                    specific? Best,Haoyu 
                  From:
                      Greg Mirsky <[e-mail]>
                      
                      Sent: Friday, August 4, 2023 1:38 PM
                      To: Haoyu Song <[e-mail]>
                      Cc: Tony Li <[e-mail]>;
                      jmh.direct <[e-mail]>;
                      Joel M. Halpern <[e-mail]>;
                      Stewart Bryant <[e-mail]>;
                      [e-mail]
                      Subject: Re: [mpls] draft-ietf-mpls-mna-hdr
                      ready to start the WGLC process
                   
                  
                    Hi
                        Haoyu,
                      I
                          consider the questions you brought up, the
                          number of NASes allowed in the stack and MNA
                          impact on MTU, to be implementation-related
                          and operational, which, in my opinion, should
                          have no impact on the ISD MNA solution. Hence,
                          let us be constructive and concentrate on the
                          questions as formulated in the original email
                      
                      
                        
                          Before
                              we ask for the RTG Dir review we'd like to
                              ask the authors if
                              there are any further changes or updates
                              that they want to do prior to
                              starting the WGLC process?
                          
                        Joel
                          provided the authors with the text addressing
                          scenario that, as I understand, we had agreed
                          requires explicit discussion in the
                          draft-ietf-mpls-mna-hdr.  Do you have an
                          additional issue that must be resolved before
                          the document is moved further in the process? 
                          I would appreciate it if your comments were
                          technical and specific.
                      
                       
                      
                      Regards,
                      
                      Greg
                      
                    
                   
                  
                    On
                        Fri, Aug 4, 2023 at 11:23 AM Haoyu Song <[e-mail]>
                        wrote:
                    
                    
                      
                        
                          I
                              don’t quite understand the point. May need
                              more explanations.  I
                              have a more general related question: Is
                              only the LER allowed to push NAS or any
                              LSR can push NAS as well? Regarding
                              to the complication, we’d better have some
                              objective method to evaluate it. My gut
                              feeling is that maintaining a single NAS
                              requires fewer lines of dataplane
                              processing code. It also keeps the header
                              overhead minimum which can potentially
                              avoid other troubles such as MTU
                              limitation. So I suggest we actually do
                              such evaluations before deciding the final
                              scheme.
                               Best
                              regards,Haoyu
                               
                            
                              From:
                                  Tony Li <[e-mail]>
                                  On Behalf Of Tony Li
                                  Sent: Thursday, August 3, 2023
                                  4:18 PM
                                  To: jmh.direct <[e-mail]>
                                  Cc: Haoyu Song <[e-mail]>;
                                  Joel M. Halpern <[e-mail]>;
                                  Greg Mirsky <[e-mail]>;
                                  Stewart Bryant <[e-mail]>;
                                  [e-mail]
                                  Subject: Re: [mpls]
                                  draft-ietf-mpls-mna-hdr ready to start
                                  the WGLC process
                              
                             
                             
                            It’s
                              more complicated and it breaks some
                              functional aspects that we have now.
                             
                            
                            Suppose
                                that the node pushing a NAS wants it to
                                have a bounded lifetime.  By placing the
                                NAS (and any copies) judiciously, it can
                                limit the effect of the NAS a portion of
                                the path.
                            
                             
                            
                            If
                                a NAS self-replicates, then this
                                capability would be lost.
                            
                             
                            
                            Tony
                            
                             
                               
                                
                                  On
                                      Aug 3, 2023, at 3:54 PM,
                                      jmh.direct <[e-mail]>
                                      wrote:
                                   
                                  
                                    
                                      That
                                          is an alternative solution. 
                                          It seems more complicated to
                                          me.
                                      
                                      Yours,
                                      
                                      Joel
                                      
                                       
                                      
                                       
                                      
                                       
                                      
                                      
                                        Sent via the Samsung Galaxy S20 FE
                                              5G, an AT&T 5G
                                              smartphone
                                        
                                      
                                       
                                      
                                       
                                      
                                      --------
                                          Original message --------
                                      
                                      From:
                                          Haoyu Song <[e-mail]>
                                          
                                      
                                      Date:
                                          8/3/23 6:41 PM (GMT-05:00)
                                          
                                      
                                      To:
                                          Tony Li <[e-mail]>,
                                          "Joel M. Halpern" <[e-mail]>
                                          
                                      
                                      Cc:
                                          Greg Mirsky <[e-mail]>,
                                          Stewart Bryant <[e-mail]>,
                                          [e-mail]
                                          
                                      
                                      Subject:
                                          RE: [mpls]
                                          draft-ietf-mpls-mna-hdr ready
                                          to start the WGLC process
                                          
                                      
                                       
                                      
                                      Then
                                          we may consider to ban the
                                          possibility of having multiple
                                          NAS replications in the stack.
                                          There will be only one and if
                                          it emerges at the top, the
                                          node must be responsible to
                                          re-push it to a proper depth.
                                          This can avoid many troubles.
                                          
                                         
                                        Haoyu
                                         
                                        
                                        
                                          From:
                                              Tony Li <[e-mail]>
                                              On Behalf Of Tony
                                              Li
                                              Sent: Thursday,
                                              August 3, 2023 3:23 PM
                                              To: Joel M. Halpern
                                              <[e-mail]>
                                              Cc: Haoyu Song <[e-mail]>;
                                              Greg Mirsky <[e-mail]>;
                                              Stewart Bryant <[e-mail]>;
                                              [e-mail]
                                              Subject: Re: [mpls]
                                              draft-ietf-mpls-mna-hdr
                                              ready to start the WGLC
                                              process
                                          
                                        
                                         
                                        
                                        
                                           
                                          
                                        It
                                          would seem that the reasonable
                                          answer is for the node pushing
                                          the NAS to understand the MSD
                                          and do the math.  This should
                                          not be hard. The rule should
                                          be simple: the NAS should fit,
                                          in its entirety, within the
                                          MSD.
                                        
                                           
                                          
                                        
                                        Now,
                                            you’re going to immediately
                                            say: what if it doesn’t? 
                                            Well, as we’ve just said:
                                            there will always be
                                            limitations.  If you have a
                                            64 byte MSD, then you
                                            probably shouldn’t be trying
                                            to use a 64 byte NAS. 
                                            There’s just not much that
                                            we can do about that.
                                        
                                        
                                           
                                          
                                        
                                        Tony
                                        
                                        
                                           
                                          
                                           
                                            
                                              On
                                                  Aug 3, 2023, at 1:51
                                                  PM, Joel Halpern <[e-mail]> wrote:
                                              
                                               
                                              
                                              
                                                A reasonable
                                                    question, separate
                                                    from the problem I
                                                    am trying to fix.Yours,Joel
                                                  On 8/3/2023
                                                      4:48 PM, Haoyu
                                                      Song wrote:
                                                  
                                                  For the
                                                      second question,
                                                      I’m asking if the
                                                      MNA substack must
                                                      be considered as a
                                                      whole or it can be
                                                      partially used.
                                                      Let’s say the
                                                      stack contains x
                                                      actions with the
                                                      depth of y. If
                                                      only x’ action
                                                      with the depth of
                                                      y’ are visible, is
                                                      that okay to only
                                                      execute these and
                                                      ignore the others?
                                                      If so, we will
                                                      have a lot of
                                                      consistency
                                                      troubles. So the
                                                      best way is to
                                                      avoid this and
                                                      ensure every
                                                      MNA-aware node to
                                                      be able to see the
                                                      entire substack.
                                                      Then what’s the
                                                      procedure and
                                                      actual limitation
                                                      of the substack
                                                      size (plus the
                                                      forwarding labels
                                                      in front of it)?
                                                      Must the node
                                                      behavior of the
                                                      entire network be
                                                      deterministic to
                                                      be able to
                                                      correctly use it?
                                                      We’d better have
                                                      clear answer to
                                                      all these
                                                      questions before
                                                      we move forward.
                                                     
                                                    Haoyu
                                                     
                                                    
                                                    From:
                                                        Greg Mirsky
                                                        <[e-mail]>
                                                        
                                                        Sent:
                                                        Thursday, August
                                                        3, 2023 1:14 PM
                                                        To: Haoyu
                                                        Song <[e-mail]>
                                                        Cc:
                                                        Stewart Bryant 
<[e-mail]>; Tony Li 
<[e-mail]>; Joel Halpern 
<[e-mail]>; [e-mail]
                                                        Subject:
                                                        Re: [mpls]
                                                        draft-ietf-mpls-mna-hdr
                                                        ready to start
                                                        the WGLC process
                                                    
                                                     
                                                    
                                                    Hi Haoyu,
                                                      I agree with
                                                          Joel, that the
                                                          question you
                                                          are asking is
                                                          likely to be
                                                          addressed as
                                                          part of
                                                          operational
                                                          procedures. I
                                                          can imagine
                                                          that MNA be
                                                          deployed in an
                                                          SDN paradigm
                                                          and that would
                                                          allow for
                                                          better control
                                                          of paths that
MNA-equipped packets take, for a more predictable actions node apply to
                                                          such packets.
                                                      
                                                      For your
                                                          second
                                                          question, I
                                                          assume that
                                                          you refer to a
                                                          situation when
                                                          ISD MNA
                                                          doesn't fit
                                                          entirely into
                                                          a cash memory
                                                          of a
                                                          particular
                                                          node. Is that,
                                                          in your
                                                          opinion, a
                                                          result of not
                                                          accurate MSD
                                                          advertisement
                                                          or a mistake
                                                          in
                                                          constructing
                                                          the label
                                                          stack and
                                                          placing ISD
                                                          MNA?
                                                      
                                                      
                                                         
                                                        
                                                      
                                                      Regards,
                                                      
                                                      Greg
                                                      
                                                    
                                                     
                                                    
                                                    
                                                      On Thu, Aug
                                                          3, 2023 at
                                                          12:03 PM Haoyu
                                                          Song <[e-mail]>
                                                          wrote:
                                                      
                                                      I have more
                                                          questions.
                                                          If we think of
                                                          a network with
                                                          mixed
                                                          MNA-aware and
                                                          MAN-unaware
                                                          nodes, the
                                                          situation
                                                          becomes more
                                                          complicated.
                                                          The
                                                          MAN-unaware
                                                          node may push
                                                          new labels
                                                          into the stack
                                                          which 
                                                          inadvertently
                                                          push the
                                                          existing MNA
                                                          substack out
                                                          of reach for
                                                          the MNA-aware
                                                          node and
                                                          effectively
                                                          make them
                                                          "unaware" of
                                                          the existing
                                                          MNA. How to
                                                          deal with
                                                          this?
                                                          
                                                          Another
                                                          question: it
                                                          that okay if a
                                                          node can only
                                                          reach a part
                                                          of the MNA
                                                          substack
                                                          (i.e., a part
                                                          of the
                                                          substack is
                                                          out of reach)?
                                                          
                                                          Best,
                                                          Haoyu
                                                          
                                                          -----Original
                                                          Message-----
                                                          From: mpls
                                                          <[e-mail]>
                                                          On Behalf Of
                                                          Stewart Bryant
                                                          Sent:
                                                          Thursday,
                                                          August 3, 2023
                                                          2:03 AM
                                                          To: Tony Li
                                                          <[e-mail]>;
                                                          Joel Halpern
                                                          <[e-mail]>
                                                          Cc: [e-mail]
                                                          Subject: Re:
                                                          [mpls]
                                                          draft-ietf-mpls-mna-hdr
                                                          ready to start
                                                          the WGLC
                                                          process
                                                          
                                                          I can see that
                                                          if a node is
                                                          going to push
                                                          some labels it
                                                          has to bring
                                                          the MNH into
                                                          view for other
                                                          nodes along
                                                          the subpath it
                                                          is adding.
                                                          
                                                          Is that dept a
                                                          network wide
                                                          parameter with
                                                          a single value
                                                          or a path
                                                          parameter and
                                                          if the path
                                                          changes for
                                                          any reason
                                                          (perhaps in a
                                                          legacy node
                                                          doing FRR)
                                                          what is the
                                                          procedure?
                                                          
                                                          Now eventually
                                                          the stack gets
                                                          popped. Is the
                                                          procedure to
                                                          copy the MNH
                                                          in its
                                                          entirety to
                                                          overwrite the
                                                          old MNH
                                                          instance? Are
                                                          there any
                                                          exceptions to
                                                          this?
                                                          
                                                          Is there a
                                                          risk that the
                                                          popping node
                                                          cannot see far
                                                          enough into
                                                          the stack to
                                                          find the new
                                                          MNH?
                                                          
                                                          - Stewart
                                                          
                                                          > On 1 Aug
                                                          2023, at
                                                          19:14, Tony Li
                                                          <[e-mail]>
                                                          wrote:
                                                          >
                                                          >
                                                          >
                                                          >> As
                                                          far as I know,
                                                          there is no
                                                          text in the
                                                          draft yet to
                                                          deal with
                                                          preserving
                                                          access to
                                                          hop-by-hop MNA
                                                          when
                                                          MNA-capable
                                                          nodes add
                                                          entries to the
                                                          label stack. 
                                                          And there is
                                                          not yet WG
                                                          agreement on
                                                          the approach
                                                          to be used, so
                                                          I have some
                                                          difficulty
                                                          proposing
                                                          text.
                                                          >
                                                          >
                                                          > I thought
                                                          that we had
                                                          agreement:
                                                          >
                                                          > If the
                                                          node is
                                                          pushing enough
                                                          labels that
                                                          the hbh NAS
                                                          would be below
                                                          the window,
                                                          then the node
                                                          needs to copy
                                                          the NAS into
                                                          the window.
                                                          >
                                                          > If the
                                                          node is
                                                          modifying the
                                                          NAS, then it
                                                          needs to make
                                                          its
                                                          modifications
                                                          and then do
                                                          the above.
                                                          >
                                                          > T
                                                          >
                                                          >
                                                          _______________________________________________
                                                          > mpls
                                                          mailing list
                                                          > [e-mail]
                                                          > https://www.ietf.org/mailman/listinfo/mpls
                                                          
_______________________________________________
                                                          mpls mailing
                                                          list
                                                          [e-mail]
                                                          https://www.ietf.org/mailman/listinfo/mpls
                                                          
_______________________________________________
                                                          mpls mailing
                                                          list
                                                          [e-mail]
                                                          https://www.ietf.org/mailman/listinfo/mpls
                                                      
                                                    
                                                  
                                                
                                              
                                            
                                          
                                           
                                          
                                        
                                      
                                    
                                  
                                
                               
                            
                          
                        
                      
                    
                  
                
              
            
          
        
      
    
  

_______________________________________________mpls mailing [e-mail]://www.ietf.org/mailman/listinfo/mpls