Abstract:  WebView Materialization for Scalable Web Servers

The frustration  of broken links from  the early Web has been replaced
today by the frustration of web servers stalling or crashing under the
heavy load of dynamically generated content. Even seemingly static web
pages  are   usually   generated  dynamically in   order   to  include
personalization and advertising  features.  Such dynamic  content  has
high resource demands and poses a scalability problem for both web and
database servers.

In my work, I have used  view materialization techniques to ameliorate
this   scalability   problem.  I  introduced    WebViews to   identify
frequently  accessed  web    page  fragments  which are    dynamically
generated.   With  materialization,  these   WebViews   are constantly
refreshed in the background in anticipation of future requests.

WebView materialization drastically improves Quality of Service (QoS),
but poses new challenges: (1) the selection of WebViews to materialize
has to be performed in real-time, with limited knowledge of the access
and update patterns, and must adapt rapidly to changing web workloads,
(2) the  Quality of Data  served  (QoD) must be  considered during the
selection process,  and, (3) scheduling  the updates in the background
concurrently with user requests has a significant impact on the QoD.

In this talk, I  will describe how to  bridge the gap between QoS  and
QoD  in scalable,  database-driven internet  servers.  I will  present
success stories for   WebView  Materialization, describe an   adaptive
Online   algorithm for WebView  selection, and   focus  on the  update
scheduling problem: determine the  order  to refresh WebViews so  that
the overall QoD is maximized.