Andre Beck wrote:
1. What is the motivation for the ordering rules in section 4? It seems to
me that there may be different business arrangements which lead to different
choices of which rule module owner takes precedence. For example, ad
insertion on behalf of access providers, ad insertion on behalf of content
providers and ad removal on behalf of clients all conflict and the
resolution depends on local context like funding models, not on a globally
fixed order.
Well, I think I have to disagree here. The client should always be able
to have the last word on any modifications he would like to make to the
requested page. [snip]
Fair enough. It is true that ultimately the client has the last say anyway (can
do
processing locally) so reflecting that precedence in the proxy is reasonable.
However, the access provider/content owner precedence may be less clear cut.
Traditionally the access provider doesn't do any semantic modification and is
transparent to the content owner. Introducing substantial, configurable, content
modification capabilities in the access network is new and it may be that
content
owners won't be fully happy with the possibilities unless they have at least the
option of some control (business level or technical).
I can also imagine cases where interleaving the rules between different
(cooperating) module owners might be useful but perhaps the complexity rules
this out.
Can you give an example where this would make sense?
The sort of thing I am thinking of here is the access provider offering service
modules which I want to invoke in my rule set, implicitly or explicitly. When
the
service is directly encapsulated as a single proxylet there is no interleaving
of
rules. However, a complex service might be implemented as a set of cooperating
proxylets tied together by a rule set. I would like to be able to invoke that
set
of proxylets as a single logical package within my client rule set, reusing the
service provider's rule set for that function.
To make this marginally more concrete ...
An "explicit invocation" example might be making my processing dependent upon a
site's privacy policy. There will be many ways of checking privacy policy (P3P,
PrivacyBank, peer-recommendation services) which may be best implemented by the
access provider as multiple proxylets rather than a single large black box.
An "implicit invocation" example is where the access provider is doing some
transcoding (e.g. image adaptation) which I want to have applied to additional
content I insert in the client rule set (I might replace ads by family photos
pulled off a remote family web site but still want the photos adapted to my
TV-viewer appliance or mobile pad by the access-network's client adaptation
proxylet).
Having said all that I'll repeat my original comment that the complexity
probably
rules this out, especially in the first pass. In the long term the division
between proxylets and the control rules is somewhat arbitrary and can probably
be
generalized. Potentially the way to go is to support rich parameterization of
proxylets rather than interleaving of rulesets.
Dave
------------------------------------------------------------------
Hewlett-Packard Laboratories | Phone: +44-117-3128165
Filton Road, Stoke Gifford | FAX: +44-117-3128925
Bristol BS34 8QZ, UK | dave_reynolds(_at_)hpl(_dot_)hp(_dot_)com