In a distributed file system the distributed storage management is made useful
to a variety of applications. Multiple quality of service options are provided
through locking. Three locking systems are provided. The system offers a locking
system designed for sequential consistency with write-back caching, typical of
distributed file systems. A second locking system is provided for sequential consistency
with no caching for applications that manage their own caches. Finally, a locking
system that implements a weaker consistency model with write-back caching, designed
for efficient replication and distribution of data is included. Locks for replication
are suitable for serving dynamic data on the Internet and other highly-concurrent
applications. The selection of the appropriate lock protocol for each file is set
using the file metadata. Further, a novel locking system is provided for the lock
system implementing a weak consistency model with write back caching. This system
is implemented utilizing two whole file locks: a producer lock P and a consumer
lock C. Any client can hold a consumer lock and when holding a consumer lock can
read data and cache data for read. The producer lock is only held by a single writer
and a writer holding a producer lock can write data, allocate and cache data for
writing. When a writer performs a write, the write is performed as an out-of-place
write. An out-of-place write writes the data to a different physical storage location
than from which it was read. By performing an out-of-place write the old data still
exists and is available to clients. Once the writer completes the write and releases
the producer lock the previous data is invalidated and the clients are informed
of the new location of the data. Clients can then read the new data from storage
when needed and the server reclaims the old data blocks.
