|
Research Members: Pedro Trancoso (graduated), Qiang Cao, and Josep Torrellas
Motivation
Database applications are a very common workload. Nevertheless,
there are very few works on analyzing the memory behavior of
this type of workload. This is becomming more important as memory
prices are decreasing. Our research interest is to study the
performance of the memory hierarchy, as well as other architectural
components, on database workloads. By this study, We want to identify the
performance bottleneck for different hardware platforms and build up
the knowledge base for the next generation processor design.
Our Work
- Our group has presented a preliminary analysis of part of the
TPC-D benchmark in the Third International Symposium on
High-Performance Computer Architecture, February 1997 .
This work, with title
The Memory Performance of DSS Commercial Workloads in
Shared-Memory Multiprocessors is available on the web.
From our preliminary work we found out that:
- Query memory behavior depends on data access method (index
or sequential scan);
- We found spatial and temporal locality in some of the
queries;
- Simple data prefetching optimization has potential for
improving the query execution.
- We characterized the TPC-D benchmark suite on a 4-way Pentium Pro
server running commercial DBMS in the International Conference on
Computer Design", October, 1999 . The paper is titled
"Detailed Characterization of a Quad Pentium Pro Server Running TPC-D".
Our conclusions of this work:
- TPC-D queries have a relatively low CPI and negligible kernel time;
- Instruction Fetch and L2 data cache misses are two major stall components;
- TPC-D, compared with TPC-C, has less kernel time, lower cache miss reates and
lower branch misprediction rates.
- We presented serveral cache-oriented optimization to enable effective
exploitation of caches in memory-resident desicsion support databases in
Cache Optimization for Memory-Resident Desicision Support Commercial Wordloads
at the International Conference on Computer Design", October, 1999 .
- These optimizations require no custom-designed hardware support and are
effective for the more complicated TPC-D queries. In a simple database, these
queries run about 13% faster with the cache-oriented optimizer and blocking,
and a total of 31% faster if, in addition, we add prefetching.
|