I think a backup solution for a non-plan9 file system (e.g. ZFS) that goes 
straight to the venti block store (as Bakul suggests) seems like it would be 
more satisfactory than layering it on top of the plan 9 file system, if there's 
a practicable way to do that.

In the Glenda and Tux paper the idea was to tunnel communication between a 
foreign file system client and server through 9P. The protocol was not really 
designed to be suitable for a storage format.

I have had occasion in the past to layer one (UNIX-like) file system on top of 
another, though: What I did was to create a single file for all the fixed sized 
metadata (inode table) and have numbered files for the inode file contents. 
Many file systems and tools don't cope well with very large directories, so I 
distributed the files over sub directories by breaking up the inode number into 
digit groups. So e.g. inumber 1234567 becomes d01/d23/d45/f67 in the underlying 
file system. Inumber 123 becomes d01/f23. Some of the files contained the 
contents of the upper file system's directories. So this way the metadata and 
path names of the two file systems are completely separated. This has the nice 
property that it uses the file implementation of the underlying file system, so 
is quite simple to implement. I didn't have ACLs or other xattr to worry about, 
but with deduplication available in the underlying file system that might not 
be a problem.
It might be reasonable to perhaps layer the storage features of ZFS over the 
Plan 9 file system in this way, though I don't know much about ZFS other than 
what I've read in wikipedia - it looks pretty intimidating! My worry is that 
(as with a lot of things) you run the risk of chasing another development team 
as they add and change things.
------------------------------------------
9fans: 9fans
Permalink: 
https://9fans.topicbox.com/groups/9fans/Taf5e3581b0a00ee0-Mbcc3476c7e5c452532c0b264
Delivery options: https://9fans.topicbox.com/groups/9fans/subscription

Reply via email to