Browse Prior Art Database

Extension of cache-control protocols in file systems to allow direct SAN data access

IP.com Disclosure Number: IPCOM000022534D
Original Publication Date: 2004-Mar-19
Included in the Prior Art Database: 2004-Mar-19
Document File: 3 page(s) / 36K

Publishing Venue

IBM

Abstract

Existing network file system protocols can be accelerated by allowing clients to access data directly across a SAN, when the existing network file system protocol has a component that controls the validity of a client cache. SAN access information can be sent from the server to the client as an attribute of the file, and the validity of that SAN access information can be guaranteed or invalidated in the same way that the protocol guarantees or invalidates the existing client data cache.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 28% of the total text.

Page 1 of 3

Extension of cache-control protocols in file systems to allow direct SAN data access

Several existing remote non-SAN file system protocols are based on delegations, wherein the server grants some logical control of an object to a client temporarily. Generally, this is coupled with a protocol by which the server may request that the client relinquish its control of the object, perhaps because the object is needed by another client or by some server-side operation. These protocols do not actually relinquish physical control of an object: the object remains the sole physical property of the server. But by protocols like this, the clients and server may share a consistent view of the object, with minimal overhead in the common cases in which objects are either read-only for many clients or read-write for only one client. The client with the delegation owns the logical view of the object in the shared space. Examples: DCE/DFS grants and revokes tokens; AFS grants "callback"s and revokes them; CIFS grants and revokes oplock (opportunistic lock); NFS version 4 grants delegations and recalls them. In all these cases, the client holding the delegation holds it until further notice from the server, or until the client itself voluntarily relinquishes it. There are fine points about what happens when client-server communication fails, in which case the server's actual physical ownership of the object allows it to decide the outcome.

But these protocols are for remote file access where file data is shipped across a network by a server. Many storage devices are today physically connected to multiple hosts via a SAN (Storage Area Network), so that any of the hosts could initiate a data transfer. Data access, for both read and write, by clients could be accelerated significantly if both file server and client were connected to the ultimate storage device with a SAN. The problem for the collection of machines is how to coordinate access to the storage device, so that both multiple clients and a server may operate on the device without interfering with each other. This article describes how existing delegation/recall protocols may be extended to cover the case of delegating not only logical control of a file, but also physical access to a file, to clients in an efficient manner.

The central idea is that non-SAN file server protocols that delegate synchronization decisions admit simple extensions by which SAN access can be delegated as well. The specifics of how the extensions can be performed depend on the specifics of the protocol. Most generally, though, a file-access delegation from server to client may be interpreted as permitting not only the possession by the client of the right to view or change the abstract shared image of a file, but also the use by the client of its direct SAN access to the storage for a file so that its data may be read or written. The server's revocation of the client's delegation may be interpreted by the client not on...