Parallel Virtual File System, Version 2
PVFS2 Development Team
September, 2003
Contents
An introduction to PVFS2
Why rewrite?
What's different?
Modular networking and storage
Structured non-contiguous data access
Flexible data distribution
Distributed metadata
Stateless servers and clients
Explicit support for concurrency
Tunable semantics
Flexible mapping from file references to servers
Tight MPI-IO coupling
Data and metadata redundancy
And more...
When will this be available?
The basics of PVFS2
Servers
Networks
Interfaces
Client-server interactions
Consistency from the client point of view
File system consistency
PVFS2 terminology
File system components
PVFS2 Objects
Handles
Handle ranges
File system IDs
PVFS2 internal I/O API terminology
Internal I/O interfaces
Job interface
Posting and testing
Test variations
Contexts
User pointers
Time outs and max idle time
PVFS2 User APIs and Semantics
UNIX I/O Interface
Permission Checking
Permissions and File Access
Access to Removed Files
Overlapping I/O Operations
Locks
MPI-IO Interface
MPI-IO Atomic Mode
MPI-IO Shared Pointer Mode
The code tree
The top level
src
src/io
test
State machines and
statecomp
Build system
Out-of-tree builds