Query Affinity in Internet Applications

Abstract:

We investigate how to improve the price-performance of Internet
application servers by caching their dynamic content in a cluster of
cache servers. In particular, we study distribution strategies that
automatically and dynamically partition and replicate data across
individual cache servers and directs queries to the right servers, in
order to maximize effective cache capacity and minimize
synchronization cost. Despite the conflicts in partitioning dynamic
content, we observe natural query affinity in a wide range of Internet
applications, which could be exploited in distribution strategies. We
have designed an affinity-based distribution (ABD) strategy for a
cluster of cache servers. We evaluate the distribution strategy with a
set of trace-based simulations. The results show that ABD reduces
cache miss ratio by a factor of 1.7 to 9 over alternative
strategies. The results also indicate that application servers,
especially those that update their back-end databases, have a higher
demand for good distribution strategies than static web servers.