Re: [gentoo-user] filesystems
On Mon, 2008-11-24 at 07:30 +0100, Dirk Heinrichs wrote: Am Sonntag, 23. November 2008 23:31:30 schrieb William Kenworthy: What I would really like is a file system that would unify these spaces and present them to the network as storage space - ideally with redundant data storage so one or more machines can dissappear and the data is still available. AFS is not quite what I want (or maybe it is, but it doesn't seem to handle transient storage duplication) For a non-native speaker, could you explain transient storage duplication a bit more? Because I think AFS may well be what you're looking for, or maybe its cousin Coda. Bye... Dirk By transient storage I mean that the data is duplicated across across physical storage spaces so that if a machine goes down, the data is still available. I thought Andrews FS did that, but didnt see when looking at it yesterday. BillK
Re: [gentoo-user] filesystems
Am Montag, 24. November 2008 11:30:25 schrieb William Kenworthy: By transient storage I mean that the data is duplicated across across physical storage spaces so that if a machine goes down, the data is still available. OK, thanks. I thought Andrews FS did that, but didnt see when looking at it yesterday. Yes, (Open-)AFS indeed does this. However, this replication is read-only. This means you can read the data as long as at least one replica is available and write the data as long as the original (the read-write) volume is available. There are also some other things to keep in mind: * AFS' primary tool for access control are its access control lists (ACL), but those are not posix, but AFS ACLs and they apply at the directory (not file) level. However, that's usually sufficient, because one can work with subdirs and symbolic links to implement more restrictive access for some files in the same directory. * ACLs can also contain host names. * If a volume is replicated, the client always prefers the read-only path (read-write volumes are usually accessed via /afs/.mycell.mydomain, while read-only volumes (if they exist) are accessed via /afs/mycell.mydomain). So if you want to modify a file you must explicitely open it via the rw-path. * Replication doesn't happen automatically, needs an explicit command. * Support for backup volumes is also there (comes with its own backup system). * Can move volumes to different servers while online. * Data is cached on the client. * You'll need Kerberos 5. If you have further questions, feel free to ask. Bye... Dirk
Re: [gentoo-user] filesystems
On Mon, 2008-11-24 at 12:07 +0100, Dirk Heinrichs wrote: Am Montag, 24. November 2008 11:30:25 schrieb William Kenworthy: By transient storage I mean that the data is duplicated across across physical storage spaces so that if a machine goes down, the data is still available. OK, thanks. I thought Andrews FS did that, but didnt see when looking at it yesterday. Yes, (Open-)AFS indeed does this. However, this replication is read-only. This means you can read the data as long as at least one replica is available and write the data as long as the original (the read-write) volume is available. There are also some other things to keep in mind: * AFS' primary tool for access control are its access control lists (ACL), but those are not posix, but AFS ACLs and they apply at the directory (not file) level. However, that's usually sufficient, because one can work with subdirs and symbolic links to implement more restrictive access for some files in the same directory. * ACLs can also contain host names. * If a volume is replicated, the client always prefers the read-only path (read-write volumes are usually accessed via /afs/.mycell.mydomain, while read-only volumes (if they exist) are accessed via /afs/mycell.mydomain). So if you want to modify a file you must explicitely open it via the rw-path. * Replication doesn't happen automatically, needs an explicit command. * Support for backup volumes is also there (comes with its own backup system). * Can move volumes to different servers while online. * Data is cached on the client. * You'll need Kerberos 5. If you have further questions, feel free to ask. Bye... Dirk Discovered this: http://en.wikipedia.org/wiki/List_of_file_systems#Distributed_file_systems Thats going to keep me busy for awhile! BillK -- William Kenworthy [EMAIL PROTECTED] Home in Perth!
Re: [gentoo-user] filesystems
Am Montag, 24. November 2008 13:03:13 schrieb William Kenworthy: Discovered this: http://en.wikipedia.org/wiki/List_of_file_systems#Distributed_file_systems Thats going to keep me busy for awhile! Interesting link. However, NFS, SMB, AFP and NCP are NOT distributed filesystems. They're just network filesystems. Bye... Dirk
Re: [gentoo-user] filesystems
On 24 Nov 2008, at 11:07, Dirk Heinrichs wrote: ... If you have further questions, feel free to ask. I would love a file system that transparently replicates over several systems - say 2 - 5. It doesn't need to amalgamate spare in any way (as BillK requests), let's just say I just have a couple of gig on each machine that I want replicated. I should be able to read operate on the files on the partition just as normal, but when a file is saved to or deleted from any one machine the change should be replicated on all the others across the (slow) network. Basically, the idea is that I should be able to set machines A, B C as MX for my domain and be able to read a new message whichever machine receives it. I should be able to run all 3 machines as IMAP servers and connect to any one of them to see the same view of my messages. When the IMAP server deletes or moves a message (on, say, A) that transaction should be replicated across B C. (But likewise if the message is moved or deleted on B then the transaction should be replicated across A C). I suspect I would be optimistic if I hoped for something so sophisticated to be readily available, as I am aware that this would be problematic to implement. But do you have any suggestions? Stroller.
Re: [gentoo-user] filesystems
On Mon, 2008-11-24 at 12:35 +, Stroller wrote: On 24 Nov 2008, at 11:07, Dirk Heinrichs wrote: ... If you have further questions, feel free to ask. I would love a file system that transparently replicates over several systems - say 2 - 5. It doesn't need to amalgamate spare in any way (as BillK requests), let's just say I just have a couple of gig on each machine that I want replicated. I should be able to read operate on the files on the partition just as normal, but when a file is saved to or deleted from any one machine the change should be replicated on all the others across the (slow) network. Basically, the idea is that I should be able to set machines A, B C as MX for my domain and be able to read a new message whichever machine receives it. I should be able to run all 3 machines as IMAP servers and connect to any one of them to see the same view of my messages. When the IMAP server deletes or moves a message (on, say, A) that transaction should be replicated across B C. (But likewise if the message is moved or deleted on B then the transaction should be replicated across A C). I suspect I would be optimistic if I hoped for something so sophisticated to be readily available, as I am aware that this would be problematic to implement. But do you have any suggestions? Stroller. I set up an openmosix cluster once using dfs I think. It replicated data just like you want so each exported thread was seeing consistent file space. It did work, but had a few issues ... I think it was designed by MS being one :) http://en.wikipedia.org/wiki/Distributed_File_System_(Microsoft) BillK
Re: [gentoo-user] filesystems
Am Montag, 24. November 2008 13:35:25 schrieb Stroller: I suspect I would be optimistic if I hoped for something so sophisticated to be readily available, as I am aware that this would be problematic to implement. But do you have any suggestions? Maybe Coda. Bye... Dirk
Re: [gentoo-user] filesystems
Am Montag, 24. November 2008 13:44:06 schrieb William Kenworthy: I set up an openmosix cluster once using dfs I think. It replicated data just like you want so each exported thread was seeing consistent file space. It did work, but had a few issues ... I think it was designed by MS being one :) http://en.wikipedia.org/wiki/Distributed_File_System_(Microsoft) I strongly doubt that you used MS DFS in a Linux based cluster. If it was really DFS, then this one: http://en.wikipedia.org/wiki/DCE_Distributed_File_System. Both are totally different beasts. Bye... Dirk
Re: [gentoo-user] filesystems
On Mon, 2008-11-24 at 13:50 +0100, Dirk Heinrichs wrote: Am Montag, 24. November 2008 13:44:06 schrieb William Kenworthy: I set up an openmosix cluster once using dfs I think. It replicated data just like you want so each exported thread was seeing consistent file space. It did work, but had a few issues ... I think it was designed by MS being one :) http://en.wikipedia.org/wiki/Distributed_File_System_(Microsoft) I strongly doubt that you used MS DFS in a Linux based cluster. If it was really DFS, then this one: http://en.wikipedia.org/wiki/DCE_Distributed_File_System. Both are totally different beasts. Bye... Dirk Your right, it was mfs - I still had the relevant line in an fstab - must have been 3-4 years ago at least. #mfs/mfsmfs dfsa=1,noauto 0 0 Its the dfsa argument that confised my memory. BillK -- William Kenworthy [EMAIL PROTECTED] Home in Perth!
Re: [gentoo-user] filesystems
Am Montag, 24. November 2008 12:07:55 schrieb Dirk Heinrichs: If you have further questions, feel free to ask. One smalll thing to add: If you decide to use it, there's a Howto under http://www.gentoo-wiki.info/OpenAFS. Do NOT use the one from gentoo.org, it's old, outdated and partly incorrect. Bye... Dirk
Re: [gentoo-user] filesystems
Am Sonntag, 23. November 2008 23:31:30 schrieb William Kenworthy: What I would really like is a file system that would unify these spaces and present them to the network as storage space - ideally with redundant data storage so one or more machines can dissappear and the data is still available. AFS is not quite what I want (or maybe it is, but it doesn't seem to handle transient storage duplication) For a non-native speaker, could you explain transient storage duplication a bit more? Because I think AFS may well be what you're looking for, or maybe its cousin Coda. Bye... Dirk