Talk:Distributed file system
|WikiProject Computing||(Rated Redirect-class)|
|The contents of the merged into Clustered file system and it now redirects there. For the contribution history and old versions of the merged article please see its history.page were|
Distributed file system
This page says "In computing, a distributed file system is a network file system where a single file system can be distributed across several physical computer nodes"
The distributed file system page says "Distributed file systems are also called network file systems. Normally many implementations have been made, they are location dependent and they have access control lists (ACLs), unless otherwise stated below."
Issue 1. These are not consistent. The first says that a distributed file system is a special case of a network file system. The second says distributed and network file systems are the same.
Issue 2. File systems are not distributed across computer nodes. File systems do not reside on computer nodes. File systems are on storage devices and those storage devices can be directly hosted by computer nodes or they may be on storage arrays. Storage arrays may be shared, i.e. they may have multiple hosts.
This Issue is the core of the distinction. Distributed file systems actually can exist across nodes or hosts (think Andrew File System) where as clustered file systems exist between nodes as the examples in that article indicate. NFS, in the most common implementation, is a rudimentary distributed file system at best. Andrews File System (AFS), and Microsoft's DFS represent more full featured examples. On the other hand, Clustered File Systems represent management of block devices in some type of shared topology (think SAN). —Preceding unsigned comment added by Sfranks (talk • contribs) 21:18, 24 May 2010 (UTC)
+1. Distributed file system is a network-accessible filesystem that is distributed ON THE SERVER SIDE, not just distributed TO the client side. Examples of distributed file systems are Andrews File System, Gluster, and even GFS. Examples of network file systems are SMB, NFS, etc. While network file systems might be distributed underneath, there is no design mechanism inherent in their protocols to enable them to do so. Thus, there should be two articles: network file system and distributed file system. To conflate the two is simply incorrect. With regards to the issue of network file system vs local (i.e., POSIX) file system: any "filesystem", whether it directly accesses disk or not, that can be accessed over the network can be considered a network file system, even if it does not fit the traditional sense of a POSIX file system. For example, FAT32 may be exported on a Windows system as SMB, and mounted on a Linux box as a samba filesystem. As long as the filesystem on the client acts like a filesystem (POSIX or otherwise) -- and, granted, this definition could get very loose (ie. MogileFS or Amazon S3), I'd personally call it a network filesystem, at least as long as it exposes commonly accepted filesystem semantics (directories, files). (S3 doesn't really expose a directory structure per se, so it'd be a stretch to call it a network file system, imo.)
Issue 3. ACLs have nothing to do with a file system being distributed, or not being distributed.
Issue 4. Is it correct to call NFS a network file system? It is a method for exporting a file system rather than being a native file system.
- Distributed file systems and network file systems are as far as I understand just two different names of the same and I believe these two articles should be merged into the distributed file system article. I found one source supporting this (Operating systems concepts by Silberschatz and Galvin ISBN 0-201-59292-4) but I do not currently have access to my Tanenbaum books right now, but his terminology is probably the one we should use.
- A distributed file system supply transparent access to the server(s) file system on the clients. One does normally say that the file system is mounted on the clients and that multiple clients access the same file system this way.
- A distributed file system, in comparison to a shared disk file system, makes it possible to have logic on the server to restrict access to the file system. It is, for example, not possible on a shared disk file system to let a single client access only that users files. I'm not really sure how this should be written on the page, but I find it important enough to mention.
- Well, Network File System (protocol) is considered a distributed file system (same source as above) and since the two terms are the same....
- --JerkerNyberg (talk) 11:18, 9 November 2008 (UTC)
I propose a merge of Network file system and Distributed file system into Distributed file system. As far as I understand the meaning of the two concepts are the same, see source above. This merge should make it more clear.
The list of distributed file systems on the page Network file system#List_of_network_file_systems could be removed (and perhaps merged) and instead have a reference to the section at the List of file systems#Distributed file systems.
I disagree that the articles should be merged. NFS is a generic, protocol-based system by which files may be shared, stored, and accessed over a network connection. It doesn't support replication or dynamic modifications. DFS, however, is a far more specific system whereby resources on multiple systems may be merged into a single, virtual directory structure. DFS requires a management interface, supports replication, and adds an intermediary layer between the file and the user that NFS does not.
- NFS was a protocol created and trademarked by SUN Microsystems with recent revisions (NFS 4 and above) managed completely through the IETF. DFS was created and trademarked by Transarc who was purchased by IBM. Dfs is the way Microsoft refers to the technology they advocate based upon IBM SMB (Microsoft adjusted to CIFS in 1996) protocol. In some of their documentation, Microsoft occasionally violates  the Transarc/IBM trademark by referring to their incompatible Dfs version as DFS. The terms for NFS (Network File System) and DFS (Distributed File System) are very clearly trademarked and the discussion about whether one is a superset of another based upon features (replication, dynamic modification, etc.) is not really relevant since they are incompatible with one another and mutually exclusive. DavidHalko (talk) 20:46, 7 February 2009 (UTC)
Is location dependent the concept you are looking for? NFS (protocol) has as far as I know been called a distributed file system. See the reference (Silberschatz, Galvin (1994). Operating System concepts, chapter 17 Distributed file systems. Addison-Wesley Publishing Company. ISBN 0-201-59292-4). Do you have any other references that use concepts the same way as you do? We need to read more books, get more references... --JerkerNyberg (talk) 23:29, 30 November 2008 (UTC)
- I think you are accurate that the SUN Microsystems NFS, IBM (formerly Transarc) DFS, Microsoft Dfs, and others may be commonly called generic network or distributed file systems. Because NFS and DFS are clearly trademarked, there is no need to reference books to resolve any questions. DavidHalko (talk) 20:46, 7 February 2009 (UTC)
The Official Samba3 HOWTO explains in chapter 20 `Hosting a Microsoft distributed file system' how the CIFS can be used to provide a distributed file system. Normal operation of a SMB/CIFS file system, however, is just as a network file system without distribution. Hence, NFS and DFS are two different things. A DFS is always a NFS, while not every NFS is also distributed. I guess the articles should not be merged, but rather amended in order to make the distinction clear. Johannes121 (talk) 15:30, 22 December 2008 (UTC)
- The SaMBa document quoted clearly identified "Microsoft distributed file system" and any reference to DFS in the opinion which followed is not accurate, since DFS is a trademark of Transarc/IBM. DavidHalko (talk) 20:46, 7 February 2009 (UTC)
If cleanup and mergers are to be done, discussion about features is not really necessary. The trademark from the original protocol creators should be used to reference Sun Microsystems "Network File System" as "NFS" and IBM's "Distributed File System" as "DFS". The Microsoft product should be referred to as "Dfs" so as not to encourage trademark infringement. I would suggest a new page called "Distributed File Systems" (note, the bold faced emphasis of the s added by me to be clear) be built as a catch-all with references to trademarked "Network File System" & "Distributed File System" and non-trademarked "Distributed File System (Microsoft)". One thing is for certain, there is more information in this talk area than in the the pages referring to the non-trademarked information and the non-trademarked information should be organized appropriately. I would be happy to do the cleanup work to make this whole topic more accurate, I could do this in a day or so. DavidHalko (talk) 20:46, 7 February 2009 (UTC)
- What a gross misjudgement. Again, Wikipedian destroys information by putting stuff together they do not know much about. --22.214.171.124 (talk) 11:30, 18 May 2010 (UTC)
See Wikipedia_talk:WikiProject_Computer_networking#Network_file_systems. -- Beland (talk) 08:56, 12 December 2009 (UTC)
I have removed banner for proposaed merge from Clustered file system. There was no corresponding banner on Clustered file system and I could find no discussion of the proposed merge. --Kvng (talk) 00:10, 24 January 2011 (UTC)
- A bot moved the discussion to Wikipedia_talk:WikiProject_Computing/Computer_networking_task_force/Archive_1#Network_file_systems. -- Beland (talk) 20:46, 7 January 2013 (UTC)
The history section says that file servers were first developed in the 1970s, and makes it more specific by mentioned the DECnet tool FAL in 1976.