Re: Why prepend metadata?

Actually we are in a much greater need of updating the object to reflect
new headers after a cache revalidation than to update the metadata. The
little metadata that needs updates is currently updated via swap.state.

Why they are in the same file is to minimize the overhead. The overhead
of the "ufs" type of stores (ufs, aufs, diskd) is already huge, and
doubling the number of files would double the overhead in both the time
needed to access the cache and space utilization (double the amount of
inodes used, and on many systems nearly double the amount of blocks
used).

Regards
Henrik

Yee Man Chan wrote:
>
> Hi
>
> Are there any good reason why we choose to prepend
> the metadata to the actual data over to put metadata
> and data in separate files? I feel that it is easier
> to update on-disk metadata in the latter case.
>
> Cheers,
> Yee Man
>
> __________________________________________________
> Do You Yahoo!?
> Yahoo! - Official partner of 2002 FIFA World Cup
> http://fifaworldcup.yahoo.com
Received on Fri May 31 2002 - 01:47:59 MDT