Comments inlined….
From:
Ken Laskey [mailto:]
Sent: 22 November 2005 15:14
To: Jones, Steve G
Cc: ;
;
Subject: Re: [soa-blueprints]
WS-TX
Metadata enables initial visibility without having to directly
interrogate the entity of interest. In fact, it helps identify the entity
as one of interest. Part of this may be an indication of what policies
will come into play and links to the details of those policies.
Where do the policies reside? Well, for different
implementations, these could reside in different places. If it is
well-known boilerplate, it may reside in some public space where it can be
easily referenced by many groups. It could also be very specific to a
given entity and held locally, but from a metadata standpoint, I would expect
the policy definition to be separate from the metadata but the metadata to
point to it. Part of policy may be how adherence to policy is evaluated,
and an evaluation may make use of a third party mechanism to ensure unbiased
tracking of compliance.
So let me get back to your specific questions:
Does the metadata reside with the service or the bus? It resides
some place where users can find it without knowing about the entity. As
for the bus, a single company may have a single bus, but I wouldn't count on a
single "universal bus".
[Jones, Steve G] Those product vendors out there will be
disappointed J But I agree federation is
going to be key (and normal)
Service has all of the policy elements and does all of the management
and adherence? I doubt whether any service will be that
self-contained. Do we really want each service to encapsulate the
evaluation function? Do we really want to try to maintain that? The
policy evaluation will be a "service" (in the business sense,
possibly accessed through one in the SOA sense) and the question is just where
that function resides. (See mention above of independent third party.)
Enterprise (the bus) that defines and enforces
policy? Policy is defined by those whose expertise is policy matters and
the enterprise will identify where the official definitions reside.
Individual services may have special policies that are coordinated with the
enterprise, but mechanisms will be needed (e.g. metadata) to indicate where
policies reside and how these are to be evaluated. Also, see note above
about non-ubiquitous USB.
[Jones, Steve G] USB = Universal Service Bus? An
almost perfect confusion point J
Service has the policy for its minimal state... bus contains the policy
related to actual interactions? It sounds like the higher level policy
defines the environment for the service and is not directly of concern to the
service. For example, in the US my income taxes are due quarterly and I
know my employer has some mechanism to get the payments in in a satisfactory
manner. I don't directly deal with this. However, it is also my
responsibility to make sure the amount sent to the IRS is sufficient and it is
my responsibility to send in any shortfall. This is my policy on how to
handle, using my mechanisms. In summary, I'll deal with policy that is my
direct responsibility and interest and I'll delegate other policies to the
appropriate "others", especially when I neither know nor care what
these policies are.
Hope this all makes sense.
[Jones, Steve G] Definitely makes sense, to me this is why
the mixed mode (and the concept of hierarchy of IT and business) should be the
norm, so policy can both be delegated upwards “You say who can call me”
and downwards “Thou shalt record everything”. So both
service container and bus must be able to exchange policy and manage policy
enforcement. I think I’m agreeing with you, I just thought it was a
good topic for debate as the “single bus” model is often pushed by
vendors, in a similar way to the “one EAI” solution was 5 years
ago.
Ken
On Nov 22, 2005, at 4:54 AM, Jones, Steve G wrote:
One question on Meta-data from a blueprints perspective, do we feel
that
Meta-data resides with the service or with the bus? This isn't an esoteric
debate, it's about how services are deployed and managed. One view is that
the service has all of the policy elements and does the all the
management
and adherence, this is the full autonomous view of the world. The other
side is that the service just provides functionality and it is the
enterprise (the bus) that defines and enforces policy. One other way I've
seen work successfully is where the service has the policy for its
minimal
state of operation, while the bus contains the policy related to actual
interactions. In this
last case the ability of the bus is only to
strengthen contracts and add them rather than being able to over-ride.
I've tended to use the later case as it fits with a hierarchy pretty
well.
But is it the blueprints going to provide examples of multiple types of
policy management approaches or stick to a single recommended best
practice?
Steve
-----Original Message-----
From: Ken Laskey [mailto:]
Sent: 21 November 2005 19:51
To: ;
; -
open.org
Subject: RE: [soa-blueprints] WS-TX
Agree. The descriptive
metadata for all entities involved in SOA,
e.g. content as well as services, should adequately describe that
entity so a user (both providers and consumers) knows what the entity
is, when it is applicable, and how to apply it. Context is obviously
a big part because the details of what is described may change with
changing context.
Ken
At 02:19 PM 11/21/2005,
wrote:
I would also add that there may be options to include some of the
work done in the WS-CAF TC for context.
It is not as rich as it should be but feedback to that group might
be worthwhile anyway.
My 2 cents.
- Dan
-----Original Message-----
From: Ken Laskey [mailto:]
Sent: Monday, November 21, 2005 10:24 AM
To: John Harby;
Subject: Re: [soa-blueprints] WS-TX
John,
I would expect the service metadata to include, or more likely
prominently link to, representations of policies under which the
service wishes to perform. Per
the SOA-RM, policies represent a
"wish list" for the service, where some items may be
negotiable and
some not. Establishing
the "execution context" is the mechanism by
which provider and consumer policies are aligned and conflicts
resolved; it may be
thought of as a negotiation of policies
resulting in a contract between the parties. A service invocation
becomes possible and may occur within an established execution
context. The transaction
semantics would not be changed, but some
level of semantic mediation may be necessary in the process of
creating the contract, including documenting what mediation has
occurred.
Ken
At 12:28 PM 11/21/2005, John Harby wrote:
I joined the WS-Transactions TC and attended their F2F last week.
Although I don't think we have the sort of liason
requirements that we may have in other SOA related TCs, the notion
of SOAs and services did come up in the discussion. One particular
example was compensation (e.g. in a rollback scenario), certain
services may be compensatable and others may not. It wasn't too
certain whether this property should be expressed in the service
metadata or some policies. It was extremely undesirable to that
group for the transaction semantics or protocol to be modified though.
--
-------------------------------------------------------------------------
--------
/
Ken
Laskey
\
| MITRE Corporation, M/S H305 phone: 703-983-7934 |
| 7515 Colshire Drive
fax:
703-983-1379 |
\
McLean VA 22102-7508
/
-------------------------------------------------------------------------
---------
--
--------------------------------------------------------------------
-------------
/
Ken
Laskey
\
| MITRE Corporation, M/S H305 phone: 703-983-7934 |
| 7515 Colshire Drive
fax:
703-983-1379 |
\
McLean VA 22102-7508
/
---------------------------------------------------------------------
-------------
This message contains information that may be privileged or
confidential and is the property of the Capgemini Group. It is intended only
for the person to whom it is addressed. If you are not the intended recipient, you are not authorized to read,
print, retain, copy, disseminate, distribute,
or use this message or any part thereof. If you receive this message in error, please notify the
sender immediately and delete all copies
of this message.
---
Ken Laskey
MITRE Corporation,
M/S H305 phone: 703-983-7934
7515 Colshire
Drive
fax: 703-983-1379
McLean VA 22102-7508
This message contains information that may be privileged or confidential and is the property of the Capgemini Group. It is intended only for the person to whom it is addressed. If you are not the intended recipient, you are not authorized to read, print, retain, copy, disseminate, distribute, or use this message or any part thereof. If you receive this message in error, please notify the sender immediately and delete all copies of this message.
