|
USENIX Technical Program - Abstract - 13th Systems Administration Conference - LISA '99
Moving Large Filesystems On-Line, Including Exiting HSM Filesystems
Vincent Cordrey, Doug Freyburger, Jordan Schwartz, and Liza Weissler, Collective Technologies
Abstract
Since the advent of Logical Volume Managers [LVM], larger individual
disk drives, and high uptime expectations, it is no longer possible at
some sites to schedule downtime windows long enough to move some very
large or very critical filesystems to new hardware. Hierarchical
Storage Management [HSM] systems share this problem. While at some
sites, users continue to enjoy the functionality of HSM based on their
specific usage patterns, sites whose usage patterns do not match HSM's
strengths have a more acute case of the same problem when moving their
data to new hardware. This paper presents a unique approach to moving
filesystems that permits a system to remain on-line and
accessible. New terminology is also introduced to assist
discussion: Forward Relocation, Reverse Relocation, and
Hybrid Relocation are defined and basic algorithms are
presented. While it is true that the total throughput rate of a
traditional dump and restore is higher, the methods presented here
require nearly zero downtime. The authors have used these
techniques to relocate data on filesystems with many small files
during the working day at several sites, as well as to exit HSM
systems as part of standard technology refresh programs. Three case
studies, where both types of data were relocated, are described in
their basic detail as successful (and ongoing) implementations. The
authors know of no prior works on this topic and hope to foster
further discussion and refinement of the techniques.
|
Need help? Use our Contacts page.
Technical Program