* Geoff Howard <[EMAIL PROTECTED]> [2004-02-24 00:31]:
> Upayavira wrote:
>
> >[changing subject...]
> >
> >Reinhard Poetz wrote:
> >
> >>From: Alan
> >>
> >>>* Geoff Howard <[EMAIL PROTECTED]> [2004-02-22 18:47]:
> >>>
> >>>
> >>>>Alan wrote:
> >>>>
> >>>>
> >>>>>* Upayavira <[EMAIL PROTECTED]> [2004-02-22 07:58]:
> >>>>>
> >>>>>
> >>>>>>I tend to think that Momento isn't suited to this need.
> >>>>>
> >>>>>>However, as an XML data repository, it seems very interesting.
> >>>>>>
> >>>>>
> >>>>>I've got a better idea of how Jisp is used in Cocoon from reading
> >>>>> all the discussion after my post.
> >>>>>
> >>>>> I suggested Momento because someone suggested Xindice which led
> >>>>> me to believe Jisp handled an XML persistence task.
> >>>>>
> >>>>> Might not be the best bet, no.
> >>>>>
> >>>>
> >>>>Still, I think finding a way to use momento to reduce
> >>>
> >>>memory overhead
> >>>
> >>>>in
> >>>>working with large xml datasets has great potential. No one really
> >>>>knows how great, but a demo/sample using it would be a
> >>>
> >>>start... (hint
> >>>
> >>>>hint :) )
> >>>>
> >>>
> >>>Working on it. As noted, I have JAXP implemented and SAX interface
> >>> to XUpdate. I have APIs. I am going to start working on services
> >>> next.
> >>
>
> JAXP... see below....
>
> >>> A Cocoon generator that takes a Momento data source and an XSLT
> >>> transform would be a start.
> >>>
> >>> I'm not sure how to get information into Momento via Cocoon. I'm
> >>> thinking about some sort of Woody binding, but that goes beyond
> >>> my current understanding of Cocoon.
> >>>
> >>
> >>speaking without following this thread closly: What about
> >>implementing a Momento source?
> >
>
> I starting to wonder if I'm being dense... wouldn't the easiest first
> test integratin be to use Memento as the JAXP xslt processor to reduce
> memory overhead on transformations of large data sets? Maybe I've
> misunderstood where/what momento is as a project? The jaxp processor
> is declared in cocoon.xconf (see instructions for switching to saxon
> for example).
I created a blog entry today:
Momento Inline
2004/02/23 12:26:54
There is a mode of operation for Momento that I've not considered
at length. Inline operation with XSLT. That is, operation where
Momento is not a data source, rather it is a transient document
object model.
This applies when performing a transform against a document that
may be too large to fit in system memory. A common use case is XML
generated from an SQL query. The SQL result set can be streamed as
a series of SAX events that clogs memory as the XSLT engine tries
to build a document representing a large data set.
There is nothing preventing Momento from building a document,
organized and clustered at the get go. More interesting would be
for Momento to build the document in memory, writing it to disk
only when memory runs out.
Currently, Momento writes its pages as it fills them. Momento might
delay a page write until a page fills, when it makes sense to do
so, but it pretty much writes the pages to disk as it writes nodes
and strings to the pages. It pools pages in memory using weak
references. When no one is writing to or reading from a page, the
weak reference will be the only reference, thus it is eligible for
garbage collection. Momento would have to intercept the garbage
collector's desire to release the page, and write it out before the
memory is released.
This means I need to develop a deeper understanding of weak
references in Java. If there is no way to hook collectection before
the fact, I'd have to rethink the paging engine so that it would
explicitly release pages as part of a MRU cache.
A hybrid of this could be used to maintain XSLT output as part of a
cache system. Momento would build the result of an XSLT transform
organized and clustered, writing it to disk when memory is tight. A
cache could use this Momento document as a source of SAX events or
as a W3 DOM document, discarding it when the upstream dependencies
change.
So, no Geoff, I'm only getting around to thinking of Momento as an
transient document object model. I'd designed it as a persistence
engine, so I'd considered Momento to be a source of data, a
place where data lives.
(I'll get back to everyone else soon. Great stuff everyone. Thank
you so very much!)
--
Alan / [EMAIL PROTECTED] / http://engrm.com/
aim/yim: alanengrm - icq: 228631855 - msn: [EMAIL PROTECTED]
