Just to be clear, I am just trying to
help think clearly through the situation before we commit to such a path
...
While the analogy to User Profile items
is tempting, I think the difference is in the area of expected stability.
User Profile data is expected to change infrequently and it is quite reasonable
for the Consumer to simply resend the entire data set when a change does
occur. I expect public parameters to contain a mixture of items where many
are relatively stable and some change relatively frequently, though hopefully
not on every user interaction (:}).
Here is my overall mental diagram:
----------- -------------------------
| Consumer | ---> | Mapping to Portlet 1
PP | -> sent to Portlet 1
| public | -------------------------
| parameter | -------------------------
| store | ---> | Mapping to Portlet
2 PP | -> sent to Portlet 2
----------- -------------------------
E.g. zip ---> P1(zipCode),
P2(postalCode)
Adding session-level caching adds some
interesting items to consider:
1. Consumers would need to track what
Portlets cache PPs for use when building SOAP messages
2. When one PP changes on a page with
many PPs, does the Consumer:
a. Resend all PPs - greatly
reduces value of doing the caching
b. Resend only changed
PPs - raises question of how to indicate it is time to use the default
value of a PP
3. Bookkeeping relative to caching:
a. If a PP changes in
such a manner that the Consumer is able to get the Portlet's markup from
a cache, the Consumer must track that the Portlet doesn't yet know about
the change.
b. Answer to Q2 impacts
whether bookkeeping is done per PP per Portlet or merely per Portlet.
Both the bookkeeping costs (memory &
processing) and value of the caching depend strongly on the answer to Q2.
Answer 2.a has less caching value, but lower bookkeeping costs.
Rich
Subbu Allamaraju <>
06/02/05 10:10 AM
To
WSRP TC <>
cc
Subject
Re: [Fwd: Re: [wsrp] Public
parameters - caching]
We have a similar situation with user profile items. These items can
change at any time depending on what goes on within the consumer's
environment. In such cases, the consumer is supposedly responsible for
resending the user profile items even when the producer stores the
parameters in the session.
Public parameters are in someways transient consumer-managed
initialization parameters. Since the portlet advertises the required
parameters at discovery time, I don't see how they can be more dynamic.
The values could change for a variety of reasons, and the consumer can
always resend when changes are detected.
Regarding Rich's comments:
1. I'm not sure if I completely follow the comment on bookkeeping.
Whether we have this kind of caching or not, the consumer will have to
have some system of storing and tracking public parameters. If the
consumer does not want to detect changes, it can always resend all
values on every request. A smarter consumer would possibly send only
those that changed from a previous request. A lazy consumer may not care
about changes during a session.
2. Stateless portlets would declare that they don't store public params
in session.
3. Whatever caching considerations we have (or don't have - I have to
check the spec) for user profile items should apply here too.
Regards,
Subbu
> I'm a bit confused now due to the fact that I missed the early
> discussions around public params. But what Subbu describes sounds
like
> some intialization params to me as they are expected to change very
> infrequent. As these are shared parameters I find the association
with
> the session not very natural as they may be invalidated and changed
by
> all kinds of external events not related to the current user.
>
> Or how would this work with the example in the spec of the zip code
if
> the user changes the zip code in some other application? So you would
> need the bookkeeping Rich mentioned, which sounds very expensive to
me.
>
>
> Stefan
>
>
> Rich Thompson wrote:
> >
> > This would change the semantics significantly, but could
also provide
> > quite a performance boost for large public parameter sets.
Effectively
> > this would change the lifetime of the parameters from a
Portlet
> > perspective to the session level ... currently they are
defined to be at
> > the request level. There would be a variety of impacts
of such a change,
> > but it definitely is an optimization to consider. Issues
I see include:
> >
> > 1. Would Consumers implement the bookkeeping needed to
track what PP
> > has changed since the last communication with each portlet
(would have
> > to take into account cache hits)?
> > 2. How would this impact stateless Portlets (i.e. no session)?
> > 3. Impacts on caching strategies where an independent cache
sits between
> > the Consumer and Portlet?
> >
> > Rich
> >
> >
> > *Subbu Allamaraju <>*
> >
> > 05/31/05 02:53 PM
> >
> >
> > To
> > wsrp <>
> > cc
> >
> > Subject
> > [wsrp] Public parameters - caching
> >
> >
> >
> >
> >
> >
> >
> >
> > Since public parameters are supplied with each
> > getMarkup/pbia/handleEvents request, isn't there a need
to let Producers
> > to store these values in the session (just like URL templates
and user
> > profile items), and let Consumers supply these values once
per session?
> > This will be a nice performance optimization.
> >
> > I did not find any references to this issue in the interfaces
SC docs.
> > Has this issue been discussed previously?
> >
> > Regards,
> >
> > Subbu
> >
> > ---------------------------------------------------------------------
> > To unsubscribe from this mail list, you must leave the
OASIS TC that
> > generates this mail. You may a link to this group
and all your TCs
> in OASIS
> > at:
> > https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
> >
> >
>
>
>
>
> ------------------------------------------------------------------------
>
> ---------------------------------------------------------------------
> To unsubscribe from this mail list, you must leave the OASIS TC that
> generates this mail. You may a link to this group and all your
TCs in OASIS
> at:
> https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
---------------------------------------------------------------------
To unsubscribe from this mail list, you must leave the OASIS TC that
generates this mail. You may a link to this group and all your TCs
in OASIS
at:
https://www.oasis-open.org/apps/org/workgroup/portal/my_workgroups.php
