NFS Version 4 Workgroup Directions
Remaining Work • NFS Version 4 Protocol – Proposed Standard – Draft Standard – Apotheosis • NFS Version 4 Implementation Choices document – Spencer
The Implementations To Do • Sun Java and Solaris (both) • Hummingbird Communications (both) • UMich OpenBSD and Linux (both) • NetApp (server) • EMC (server) • OpenBSD by Rick Macklem (no bake- off participation yet)
Implementations and IETF • To move from Proposed to Draft Standard – Minimum six months – Two independent interoperating implementations required • Gated on IESG to Proposed Standard • Question of the day: When is it safe to ship?
Regarding Number of Implementations • There are seven implementations in progress – Rick Macklem is pursuing an eighth in OpenBSD • There were only two implementations complete on V3 protocol freeze • Version 4 is doing great – Implementation was not require to achieve Proposed Standard – so we are ahead of the game
When “to ship”? • IETF guidelines say Draft Standard is stable (and implies shippable) • Proposed Standard is risky – Subject to change following implementation experience • Linux and OpenBSD source will be widely available before Draft Standard – Issue of how to control risk of old non-compliant implementation proliferation
Future Work • Migration/replication • SNMP MIB • Name spaces/LDAP schemas • Service Location Protocol • Disconnected operation • Performance issues in protocol • Proxy cache support
Basics • Not all work will necessarily involve protocol changes – Name space definition may be described via practices and informational RFC’s for example • Minor revisioning support exists in NFS Version 4
Migration and Replication • Migration is the movement of a “file system” from one server to another – Useful only when transparent – Necessary even when not • Replication is the creation of one or more copies of a “file system”
NFS Version 4 “file system” • Has a file system ID • A closed set of unique “file ids” – aka inodes ☺ • A set of attributes associated with the “file system” • The basis for replication and migration?
Sticky issues • fsid should define a “file system” – Hard links exist within the file system
Migration/Replication • NFS Version 4 defines client to server interaction – The volatile file handles allow cheesy solutions ☺ – Hashed file names persist ☺ – Undefined is server-to-server protocol • Prior work for server-server – AFS/DFS – rdist – rsync
Issues around M/R • Efficient updates – Minimize client suspension – Minimize transfer/updates • Atomicity guarantees • Replica integrity • Read-only vs. write replication – Write replication is harder, but…
Issues in M/R • File level protocol vs. block level protocol – Multi-platform support
Practically speaking • Re-charter working group to pursue migration/replication (and other things) • Charter new working group (please no) • Timing? – Don’t want to drain resources to complete core priorities
SNMP MIB • Just work • Informal, non-standard MIBs done in past for Version 2 and 3? • SNMP has not necessarily been synonymous with mgmt – Other avenues to pursue
Name spaces/LDAP schemas • The “global” name space issue – Service location? • Implementation document to define standards for naming? – Practice best backed by mechanism to support it
Finding servers • Finding resources (servers and file systems) in the network • Windows does this, AFS/DFS does this
Disconnected operation • Coda and Windows 2000+ play in this space • Client implementation issue primarily? • Minor protocol revision (error codes, etc) for reconciliation – Similar issues at 10,000 ft to multiple replica consistency and reconciliation
Performance • NFS Version 4 involves more complex marshalling and unmarshalling – Coupled to more complicated error handling due to COMPUND operation – Implementations not sufficiently progressed to measure hit • Protocol offload – Sapuntzakis comments – Minor versioning for protocol tweaks?
Tuesday 1:00 – 2:00 • To discuss further migration and replication • Other futures
Follow up • Contact area directors for guidance on future work (distraction, focus, re-charter, etc.) • Produce strawman for migration/replication • Produce strawman for SNMP MIB • Determine if other mgmt methods need investigation • Contact working group alias to determine interest in future work
Recommend
More recommend
