Re: [ossig] Two obscure questions: directory access times and atomic rename()

[Julian Gomez <kluivert@xxxxxxxxx>]

On Wed, Sep 08, 2004 at 10:56:54AM +0800, Christopher DeMarco spoke thusly:
>On Tue, Sep 07, 2004 at 07:19:23PM +0800, Julian Gomez wrote:
>
>> Break up their queue structure.
>
>Yup.  But  that's a useless fix  by itself, as  without a check on the
>number of inodes allowed to exist in a given sub-directory you'll just
>wind up with a BUNCH  of really obscenely  stuffed-full queuedirs.  So
>my  question was whether there's   a "magic  number" of inodes  beyond
>which directory   access  slows  worse-than-linearly,  or an  abstract
>method  for determining that  number   (aside, of course, from  simply
>waiting until you hear/smell your disks thrashing).

I don't have a magic figure for you. My laptop moans when opening 200++ :)

If you google for "mbox vs maildir comparisons" I think you might find some
helpful hints.

The queue structure break-up for MTAs work because the files in the queue
are mostly all short-lived anyhow, pending final destination delivery.
YMMV.

>> Dinesh  noted, its due   to the linear  lookup table  implementation
>> which doesn't scale particularly well.
>
>Is this implementation fs-specific or UN*X flavor-specific?

filesystem specific.


---------------------------------------------------------
To unsubscribe: send mail to ossig-request@mncc.com.my
with "unsubscribe ossig" in the body of the message