4th USENIX Conference on File and Storage TechnologiesAbstract
Pp. 115128 of the Proceedings
Second-Tier Cache Management Using Write Hints
Xuhui Li, Ashraf Aboulnaga, and Kenneth Salem, University of Waterloo; Aamer Sachedina, IBM Toronto Lab; Shaobo Gao, University of Waterloo
Storage servers, as well as storage clients, typically have large memories in which they cache data blocks. This
creates a two-tier cache hierarchy in which the presence
of a first-tier cache (at the storage client) makes it more
difficult to manage the second-tier cache (at the storage
server). Many techniques have been proposed for improving
the management of second-tier caches, but none
of these techniques use the information that is provided
by writes of data blocks from the first tier to help manage
the second-tier cache. In this paper, we illustrate how
the information contained in writes from the first tier can
be used to improve the performance of the second-tier
cache. In particular, we argue that there are very different
reasons why storage clients write data blocks to storage
servers (e.g., cleaning dirty blocks vs. limiting the time
to recover from failure). These different types of writes
can provide strong indications about the current state and
future access patterns of a first-tier cache, which can help
in managing the second-tier cache. We propose that storage
clients inform the storage servers about the types of
writes that they perform by passing write hints. These
write hints can then be used by the server to manage the
second-tier cache. We focus on the common and important
case in which the storage client is a database system
running a transactional (OLTP) workload. We describe,
for this case, the different types of write hints that can
be passed to the storage server, and we present several
cache management policies that rely on these write hints.
We demonstrate using trace driven simulations that these
simple and inexpensive write hints can significantly improve
the performance of the second-tier cache.
- View the full text of this paper in HTML and PDF.
Until December 2006, you will need your USENIX membership identification in order to access the full papers. The Proceedings are published as a collective work, © 2005 by the USENIX Association. All Rights Reserved. Rights to individual papers remain with the author or the author's employer. Permission is granted for the noncommercial reproduction of the complete work for educational or research purposes. USENIX acknowledges all trademarks within this paper.
- If you need the latest Adobe Acrobat Reader, you can download it from Adobe's site.