[OpenAFS] Re: why afs backup is so poorly supported

Steve Simmons scs@umich.edu
Thu, 12 Oct 2006 08:58:14 -0400


On Oct 10, 2006, at 5:24 PM, Matt Benjamin wrote:

> Cool as the transactional piece would be, iirc from our discussion  
> in 2004, putting a postgresql behind every fileserver sounds kind  
> of heavyweight, doesn't it?
>
> Plus, is there a difference between transactional metadata updates  
> and transactional file data updates?

There have been various claims about the overhead of using a dbms as  
a file server. There's an excellent paper out there somewhere  
(translation: I'm too damned lazy to look for it (again)) on the  
Inversions file system, a project which built a fs on top of  
University Ingres. It was presented at a USENIX.  As I recall, they  
claimed somewhere about a 2:1 overhead in general.

With respect to the metadata differences, yeah, they're huge. Most of  
what a file system would call metadata is just another entry in a  
table. Those tables themselves have metadata, but if an entire fs  
worth of metadata were contained in a single table, the transactional  
costs would be hugely smaller (in my humble and only semi-educated  
opinion, of course).