On 1/7/06, Avishay Traeger <atraeger@xxxxxxxxxxxxx> wrote:
thank you for clarifying the concept of stackable file systems which i was unaware of .. i was talking about the RAID like functionality only .. i.e. for doing striping or mirroring .. in this case you have to put your logic of stripping or mirroring in the code of the RAIF_read which will do the necessary calculations like strip size calculations on each branch then will issue the underlying file system read function for each strip i.e will implement your functionality before doing the underlying file system calls
sorry if i couldn't understand your problem .. but i've answered your question the way i've solved my problem for implementing RAID functionality on a cluster where each node of the cluster has it's own file system .. which i suppose it's the same problem with your own if you look at it with some abstraction
Regards
MHD.Tayseer
> On Sat, 2006-01-07 at 11:22 +0530, prasad Musale wrote:
> > Hello Friends,
> > I am doing work on stackable file system RAIF, which
> implements the
> > RAID functionality at file system level. I got confused
> about where to
> > put my stackable functions. E.g. for reading a file sys_read
> calls the
> > vfs_read(), which in turn calls the lower level fs read()
> function(In
> > most of the fs it's generic_file_read()).
> Generic_file_read() uses
> > page cache & get page from disk .Now if I want to add my
> > functionality(Stripping or mirroring), where should I add
> it?
> > In generic_file_read() or before generic_file_read(that
> means
> > capturing the pointer from vfs, adding our function & then
> calling
> > generic_file_read.)?
> > Does anyone knows more about stackable fs. I know the basic
> design but
> > do not know it at coding level.
> > Thanks.
It doesn't seem like you are familiar with stackable file systems. To
put it very simply, in this environment you will have multiple versions
of each file system object (for most objects) instead of one. Let's say
you have RAIF mirrored with four branches, where each branch is an ext2
file system. For each file, you will actually have five inodes - one
RAIF inode, and four ext2 inodes. The VFS will call functions from the
RAIF inode's operation vector. This function should then call
operations from the lower-level inode operation vectors. So for
example, the RAIF read function should call one of the lower level
inodes to do the read if it is mirrored, but should read from all lower
level inodes if it is striped.
thank you for clarifying the concept of stackable file systems which i was unaware of .. i was talking about the RAID like functionality only .. i.e. for doing striping or mirroring .. in this case you have to put your logic of stripping or mirroring in the code of the RAIF_read which will do the necessary calculations like strip size calculations on each branch then will issue the underlying file system read function for each strip i.e will implement your functionality before doing the underlying file system calls
sorry if i couldn't understand your problem .. but i've answered your question the way i've solved my problem for implementing RAID functionality on a cluster where each node of the cluster has it's own file system .. which i suppose it's the same problem with your own if you look at it with some abstraction
Regards
MHD.Tayseer
It seems like you are not familiar with the key concepts here - you
should definitely read the following papers:
http://www.fsl.cs.sunysb.edu/docs/raif-ccgrid05/raif.html
http://www.fsl.cs.sunysb.edu/docs/linux-stacking/index.html
http://www.fsl.cs.sunysb.edu/docs/wrapfs/index.html
Good luck with the project!
Avishay Traeger
http://www.fsl.cs.sunysb.edu/~avishay/
--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/
