|
WebViews
The World Wide Web has seen tremendous growth in the years since its
inception, accompanied by a big transformation in its nature, from
purely static HTML pages in the early 90s to most web pages having
some dynamic content today. Online services, frequently updated
content and personalization are the main reasons behind dynamically
generated web pages. On the one hand, dynamic content requires far
greater resources from web servers than static pages do and does not
scale well. On the other hand, an increasingly big fraction of the
world population relies on online services to perform everyday tasks,
from reading the newspaper and shopping, to looking up movie times and
real-time stock information.
Our research focuses on improving the performance of database-backed
web servers which are commonly used to generate dynamic content on the
Web today. The ultimate goal would be to solve the scalability issue
(i.e. allowing the web server to handle a great number of users),
without sacrificing the ``quality'' of the served information
(i.e. always responding with ``fresh'' data).
Although web caching has solved the scalability problem for static
pages, it cannot be directly applied for dynamically generated pages,
since it will not provide any guarantees for the freshness of the
cached data. Servicing user requests fast is of paramount importance,
only if the data is fresh and correct, otherwise it may be more
harmful than slow or even no data service. We have already showed
through experiments on an industrial-strength prototype that web
materialization, where web pages are cached and constantly
refreshed in the background, is a robust solution to the scalability
problem for dynamic web content. We use the term WebView
to refer to the HTML fragments that are the unit of materialization.
Similarly to traditional database views, WebViews can be in two forms:
virtual or materialized. Virtual WebViews are computed dynamically
on-demand, whereas materialized WebViews are precomputed. In the
virtual case, the cost to compute the WebView increases the time it
takes the web server to service the access request (the query response
time). On the other hand, in the materialized case, every update to
base data leads to an update to the WebView, which increases the
server load. Having a WebView materialized can potentially give
significantly lower query response times, compared to the virtual
approach. However, it may also lead to performance degradation, if the
update workload is too high.
WebView materialization generates the following issues, which we are
addressing in this project:
WebViews Papers
WebViews People |
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
|
|
|||||
