Body vs Header Correlation - which is best?

We've been talking about whether or not we should specify how to do
correlation, so I thought it would be a good idea to suggest some
alternative ways of doing correlation together with their "pros and cons" so
that we can, hopefully, make some type of objective decision on what, if
anything, to do.
 
Note that this problem does not apply just to correllation, but also how you
identify the choreography type being performed, and the specific message
(interaction) within a choreography type that a message represents -
although I don't always mention this specifically in the descriptions below.
 
So here are the five options I've thought of ...
 
HEADER ONLY CORRELATION
Put the correlation, choreography type, etc, in a header on the message
(e.g. a SOAP header)
Advantages:
1. Independent of the payload - works with any message format
2. Works with message payloads that do not contain any data that can be used
for correlation, etc
3. Independent of the application that is processing the payload
4. Simple correlation mechanism
Disadvantages:
1. If there is data in the payload that can also be used, then it might be
inconsistent
2. Correlations might not be simple if there are many overlapping threads.
 
BODY ONLY CORRELATION
Use information for correlation, choreography type, etc. contained in the
body of a message, e.g. an order Id in an order document.
Advantages:
1. Re-uses "real" often, business derived, data.
2. No duplication or inconsistency possible
Disadvantages:
1. Requires that the content and structure of the payload is understood
before correlation, etc. can be done or checked
2. Ties correlation to a particular message format - means generalized
correlation mechanisms can't be used
3. Requires the application does the correlation, choreography type checking
etc, instead of payload-independent middleware doing instead
 
BODY WITH HEADER REFERENCES
Use information in the body (as in Body Only Correlation) but also provide
"pointers" of some kind in a header (e.g. a SOAP header) which points to the
where within the body the correlation information is contained. This is, I
believe, the BPEL suggested approach.
Advantages:
1. Potential for a generalized approach - payload-independent software can
follow the references to find the data being referenced
2. Removes the need for the application to do correlation
Disadvantages:
1. The different types and formats of the data being referenced and the
payloads being used will significantly adds to the complexity of following
message paths.
2. Probably restricted to XML payloads. For example if the payload was a PDF
file then this would not work (no flames please for not requiring use of XML
! )
3. Possible performance problems - especially if the message payloads are
large.
4. Only works if the message is not encrypted. If the payload is encrypted
then it might not be possible to resolve the reference
 
BODY WITH HEADER COPY
Use information in the body (as in Body Only Correlation) but this time
*copy* the information from the body into a SOAP header.
Advantages:
1. Independent of the message format
2. Removes the need for the application to do the correlation
3. No performance problems as payload does not need to be examined
4. The payload can be encrypted and only the data required for correlation
copied to the header.
Disadvantage:
1. Probably need to map the payload data (e.g. integer) to a standard
external one in the header (e.g. string)
2. Duplicate data with potential for inconsistencies between the body and
header versions of the correlation data
 
BODY AND HEADER SEPARATE
Use information in the body, but with separate, independently created,
correlation, etc id's in the header
Advantages:
1. Independent of the message format
2. No need for the application to do the correlation
3. No need to map between body and header formats
4. Same approach whether or not the body contains data that can be used for
correlation
5. Header correlation information need bear no resemblance to the payload
equivalent which may be encrypted
6. No performance problems as payload does not need to be examined
Disadvantages:
1. Potential for inconsistencies between the header and body mechanisms for
correlation
2. Does correlation twice in two different ways - in the header AND in the
body.
 
I hope I've thought of all the options - are there any more?
 
I also hope I've covered all the advantages/disadvantages fairly - any
comments?
 
Now which is aprpoach is "best"? One of them? Some of them? All of them !!!
 
Thoughts anyone ... I'll save mine for later ...
 
David
PS Should I copy this to the BPEL list?

Director, Standards Strategy
Commerce One
4440 Rosewood Drive, Pleasanton, CA 94588, USA
Tel/VMail: +1 (925) 520 4422; Cell: +1 (925) 216 7704
< mailto:david.burdett@commerceone.com
<mailto:david.burdett@commerceone.com> >; Web: < http://www.commerceone.com
<http://www.commerceone.com/> >