Capwap Mailing List: Re: Proposed Resolution for Issue 224/89 (and part of 146)

[Jim Murphy (jmurphy]

Bob,

I see no reason why this optimization can not be
used in the control channel as well. If so, then the
concerns about the data and control channel having to
be exactly the same are eliminated with the added benefit
of improved performance in each case.

That said, I think we could improve the format a bit based on
feedback from Sudhanshu and some new observations based on the
idea of packet header overlay.

The basic idea is that Version and Type fields are what is needed
for the "preamble" and exist for every type of packet. However, only
8 bits are required. We then define the DTLS shim and CAPWAP Header to
include the preamble. The resulting format is something like this:

CAPWAP Preamble
        0
        0 1 2 3 4 5 6 7
       +-+-+-+-+-+-+-+-+
       |Version| Type  |
       +-+-+-+-+-+-+-+-+

   Version:  A 4 bit field which contains the version of CAPWAP used in
      this packet.  The value for this draft is zero (0).

   Payload Type:  A 4 bit field which specifies the payload type that
      follows the preamble header.  Unsupported values MUST be silently
      dropped. The following values are supported:

      0 -  Clear text.  If the packet is received on the data UDP port,
         the CAPWAP stack MUST treat this as a clear text CAPWAP data
         packet.  If received on the control UDP port, the CAPWAP stack
         MUST treat this as a clear text CAPWAP control packet.  If the
         control packet is not a Discovery Request or Response packet,
         it is illegal and MUST be dropped.

      1 -  DTLS Payload.  The packet is either a DTLS packet and MAY be
         a data or control packet, based on the UDP port it was received
         on (see section Section 3.1).

CAPWAP DTLS shim:

        0                   1                   2                   3
        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
       |Version| Type  |                    Reserved                   |
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Or, said slightly differently:

CAPWAP DTLS shim:

        0                   1                   2                   3
        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
       | Preamble      |                    Reserved                   |
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

For the header we then have the following:

CAPWAP Header:

        0                   1                   2                   3
        0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
       |Version| Type  |  HLEN   |  RID    |  WBID   |T|F|L|W|M|K|Flags|
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
       |          Fragment ID          |     Frag Offset         |Rsvd |
       +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+


Note, Version/Type are the CAPWAP preamble.

Essentially, what we have created a de-multiplexing scheme that has
the following properties:

 - Its performance characteristics are no worse than those of the -04
   preamble. In fact, demuxing directly off the preamble type is
   very attractive from a performance perspective.
 - It does not waste 32 bits in the clear path case.
 - It works equally well for the data path and the control path so
   it is a universal solution.
 - It provides for a powerful extensibility feature that will enhance
   the longevity of the protocol.

Other notes:

 - I have transposed RID and HLEN to give the hint of a TLV structure.
   I think this is a useful concept that we should support in the
   protocol.

 - I belive the F and L flags could be moved to the Rsvd area of the
   fragmentation super frame as suggested by David Perkins. This opens
   up some more flag space in the first super frame.

Thanks,

Jim

Bob O'Hara (boohara) wrote:
>  
> It's always fun to be part of an exercise to optimize something until it
> can't be optimized any further, and let's be clear about it.  That is
> what we are doing here.  The current preamble, common to both control
> and data packets, works.  What is being proposed is to make the data
> packet preamble as short as possible, by reducing its size by 32 bits.
> This comes at the cost of having the control and data packet formats
> diverge.
>
> Let me propose some reasons for keeping the preamble of the control and
> data packets the same as they were.
>
> 1. A DTLS-protected packet, either control or data, is handled exactly
> the same way to produce the decrypted CAPWAP payload.  This decrypted
> payload can then be passed to software for processing (either control or
> data) or can be passed to fast path hardware for data path processing.
>
> 2. 32 bits take exactly 32ns to transmit at a gigabit per second, which
> is likely to be the predominant connection for both WTPs and ACs.
> Optimizing the protocol to save these 32ns is a foolish economy.  Is
> there a dire cost that we encounter, in order to send these bits?
>
> 3. Having two different CAPWAP preambles doubles the cost of development
> of this portion of the protocol (particularly if the CAPWAP header
> cracking is done in hardware), doubles the hardware necessary to process
> this portion of the packet (perhaps even that necessary to process the
> entire packet), and doubles the number of bugs to discover and fix.
>
> I believe these practical reasons outweigh the reasons presented for
> making the change to the header.
>
>  -Bob
>  
> -----Original Message-----
> From: Jim Murphy [mailto:jmurphy [at] trapezenetworks.com] 
> Sent: Thursday, January 25, 2007 12:19 AM
> To: capwap [at] frascone.com
> Subject: Re: [Capwap] Proposed Resolution for Issue 224/89 (and part of
> 146)
>
>
> Please consider the following alternative proposal to optimize the data
> channel when no DTLS encryption is present. With this proposal,
> CAPWAP data channels running in the clear will not require the CAPWAP
> preamble. However, CAPWAP data channels running DTLS must have the
> CAPWAP preamble.
>
> The CAPWAP preamble is modified as follows:
>
> [...]
>          0                   1                   2                   3
>          0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
>  
> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>         |Version| Type  |                    Reserved
> |P|
>  
> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
> [...]
>      P: Must be 1. Indicates that this is a CAPWAP preamble.
> [...]
>
> The CAPWAP Header is modified as follows:
>
> [...]
>          0                   1                   2                   3
>          0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
>  
> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
>         |Version|   RID   |  HLEN   |  WBID   |T|F|L|W|M|K|    Flags
> |P|
>  
> +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
> [...]
>      P: Must be 0. Indicates that this is a CAPWAP preamble.
> [...]
>
> The basic idea is to overlay the CAPWAP preamble and the first
> 32 bits of the CAPWAP Header. Note the intentional overlap of
> the Version field and the P bit. Essentially the P bit is a type
> indicator that indicates the type of super field present. A 1
> indicates a CAPWAP preamble, a 0 indicates the first 32 bits of
> the CAPWAP Header.
>
> Any data packet on a clear (unencrypted) data channel looks
> as follows (to illustrate the use of the P bit):
>
>         CAPWAP Plain Text Data Packet:
>         +--------------------------------+
>         | IP  | UDP | CAPWAP | Wireless  |
>         | Hdr | Hdr | Header | Payload   |
>         |     |     | P=0    |           |
>         +--------------------------------+
>
> Any data packet on an encrypted data channel or a DTLS
> session establishment packet looks as follows:
>
>          DTLS Secured CAPWAP Data Packet:
>         +------------------------------------------------------+
>         | IP  | UDP | CAPWAP | DTLS | CAPWAP | Wireless | DTLS |
>         | Hdr | Hdr | p-amble| Hdr  |  Hdr   | Payload  | Trlr |
>         |     |     | P=1    |      |        |          |      |
>         +------------------------------------------------------+
>                               \----- authenticated -----/
>                                     \------- encrypted --------/
>
> A switching entity need only check the CAPWAP Version and then the P
> bit to determine if the CAPWAP packet needs DTLS processing.
> If the P but is not set, the switching entity may immediately
> assume only a CAPWAP header and commences de-encapsulation and
> possible reassembly processing.
>
> This proposal serves the following purposes:
>
>   - The CAPWAP preamble is present only when really needed. Specifically
>     to identify CAPWAP packet attributes outside of the DTLS encrypted/
>     authenticated area when DTLS is used.
>
>   - Eliminates the waste of 32 bits of header information to convey a
>     single bit of information when in the clear.
>
>   - Allows for the continued use of the CAPWAP preamble for other
>     purposes, such as DTLS session de-multiplexing to deal with the
>     issue of QoS reordering. (see earlier email from Mani - The QoS DTLS
>     factor)
>
> Please let me know if you have any questions.
>
> Thanks,
>
> Jim
>
> _________________________________________________________________
> To unsubscribe or modify your subscription options, please visit:
> http://lists.frascone.com/mailman/listinfo/capwap
>
> Archives: http://lists.frascone.com/pipermail/capwap
> _________________________________________________________________
> To unsubscribe or modify your subscription options, please visit:
> http://lists.frascone.com/mailman/listinfo/capwap
>
> Archives: http://lists.frascone.com/pipermail/capwap