|
USENIX Technical Program - Abstract - WinsSys - August 2000
Archipelago: An Island-Based File System For Highly Available And
Scalable Internet Services
Minwen Ji, Edward W. Felten, Randolph Wang, and Jaswinder Pal Singh, Princeton University
Abstract
Maintaining availability in the face of failures is a
critical requirement for Internet services.
Existing approaches in cluster-based data storage rely on redundancy to
survive a small number of failures, but the system becomes entirely unavailable
if more failures occur. We describe an
approach that allows a cluster file server to isolate failures so that the system can continue to serve most
clients. Our approach is complementary
to existing redundancy-based methods: redundancy can mask the first few
failures, and failure isolation can take over and maintain availability for the
majority of clients if more failures occur.
The building blocks of our design are self-contained and
load-balanced file servers called islands.
The main idea underlying island-based design is the one-island principle: as
many operations as possible should involve exactly one island. The one-island
principle provides failure isolation because each island can function
independently of other islands' failures. It also helps the file system scale
with the system and workload sizes because communication and synchronization
across islands are reduced. We implemented a prototype island-based file system
called Archipelago on a cluster of
PCs running Windows NT 4.0 connected by Ethernet. The measurement of micro
benchmark shows that Archipelago adds little overhead to NTFS and Win32 RPC
performance; while the measurement of operation mixes based on NTFS traces
shows a speedup of 15.7 on 16 islands.
- View the full text of this paper in
HTML form and
PDF form.
- If you need the latest Adobe Acrobat Reader, you can download it
from Adobe's
site.
- To become a USENIX Member, please see our Membership Information.
- Current USENIX Members may change their password.
|
Need help? Use our Contacts page.
