[OpenAFS] Re: [possibly dumb question] volume must occupy entire OS-level filesystem?

Brandon S. Allbery KF8NH allbery@ece.cmu.edu
Sun, 18 Dec 2005 21:07:04 -0500

On Dec 18, 2005, at 8:47 , zeroguy wrote:

> On Sun, 18 Dec 2005 17:02:10 -0800
> Adam Megacz <megacz@cs.berkeley.edu> wrote:
>> Regarding the paragraph above, I know what inodes are and the  
>> point of
>> the namei() system call, but I wasn't aware that AFS fileserver
>> instances came in two "flavors" with these names, or their exact
>> meaning in this particular context (though I can take a guess).
> Unless I'm mistaken, fewer people are using inode-based fileservers
> these days, as the benefits of running one over a namei server  
> (speed?)
> are becoming less significant, and it's supported on fewer OSes (I
> think...)

inode fileservers need to know far more about the internals of the  
host filesystem, which makes them much less portable; namei  
fileservers can run on top of almost anything.  inode is slightly  
faster, which mattered when the average fileserver was a pmax or etc.  
but is much less relevant on modern, or even 2 years back, hardware.

Additionally, fsck on an inode fileserver partition will destroy it  
since all the volumes are raw inodes not attached to the filesystem  
with magic metadata which fsck thinks is erroneous (IBM AFS used to  
ship with "vfsck" which understood this and handled it correctly, but  
it had to be based on vendor code that wasn't always available to  
OpenAFS); but normal fsck is safe on a namei partition.

brandon s. allbery     [linux,solaris,freebsd,perl]       
system administrator  [openafs,heimdal,too many hats]   
electrical and computer engineering, carnegie mellon university