Damelo! An Explicitly Co-locating Web Cache File System
Author: Jonathan Ledlie
Contact: ledlie@cs.wisc.edu
How to best relate cached web objects is a complex and on-going problem
and always depends on the workload. By locating objects which are likely
to be referenced together adjacently on disk and then importing them
into memory as a group, large reductions in user-perceived read latency
are possible. Instead of discovering new policies for relating objects,
we have built a specialized file system called Damelo which leverages
its data's cache-only nature for speed while leaving the determination
of how to relate objects as a flexible policy external to the system.
By providing a simple but effective interface for co-location,
Damelo provides much higher performance for its specialized web object
workload than a standard Unix file system. The purpose of this thesis
is to design, implement, and test the Damelo file system with
microbenchmarks and real workloads and to build a multithreaded
networking layer to propel web objects through it.
Project
@mastersthesis{ ledlie00damelothesis,
author = "Jonathan Ledlie",
title = "Damelo! An explicitly Co-locating Web Cache File System",
school = "University of Wisconsin",
year = "2000"
}
Related Papers
Resources
Results
Summary of results, showing that the average read latency for small files is significantly better for Damelo(D) than for a web cache running on Linux's ext2 file system.
Last update:
