At 17:09 2001-02-16, Geir wrote:
>I am very interested to see how that turns out in practice. I have a
>client who went down that road, providing content from the DB (Oracle),
>and quite frankly, it doesn't work performance-wise. DBs just don't
>seem to be great at pumping bitstreams, no matter what Bill and Larry
>tell you. Finding stuff, sorting stuff, storing suff yes....
Yup, if every request has to go to the database it will surely be
a significant bottleneck. But, as you say in the next paragraph:
>The result is a tiered data architecture where the data is maintained in
>the database, but never delivered from there - when something changes,
>it is rendered out of the DB into the form needed (usually HTML files),
>and cached simply as a file on disk. All requests are fed the file
>version, and it is only changed when the content changes in the db.
Indeed. What I neglected to mention in my previous post is the fact
that we employ very aggressive caching on several levels in the
application, i.e. url matching to sites, page lookups, data and
content generation and so on. What it typically leads to is that
>90% of all requests results in cached data being delivered back
to the client. A few specific parts of the application need to
run uncached, obviously, but they are mainly the management parts,
i.e. the site and page management modules. The majority of the others
can without problems run with cached data for 1, 5 or even 60 minute
intervals.
The modules themselves, where a module can be anything from a clock
displaying current date and time to a XML-based content management
mini-application, are free to implement whatever caching mechanisms
are necessary to ensure optimal performance without sacrificing
timeliness in the delivered content. Caching can be done in memory
or to the file system if the amount of data calls for it.
With 4-5 completely uncached modules in a page and two layers of
templates (yep, the modules use templates as does the encapsulating
page) my puny little development machine manages to crank out well
over 100 pages/second when stresstested. Things look promising so far.
> >
> > Other loaders: The ServletLoader sounds cool and could definitely
> > be an alternative solution for our system. Didn't think of that
> > approach before. Cool idea!
>
>JAR! JAR!
>
>JarJarLoader
BinksLoader? Ehh... :-P
--
David Kinnvall
Alert Investor Relations AB
+46 31 802640/10, +46 31 802670 (fax), +46 705 818432 (mobile)
