Re: mbox vs. Maildir (was: Re: [Evolution] expunge speed)

On 21 Aug 2001 00:19:39 +0200, Alexander Skwar wrote:

Size?  Well, according to 'du' the Maildir is bigger than the single
file.  Which GOTTA be wrong.

no - it's almost certainly right.  remember that disks are not allocated
bit-by-bit - they're allocated in blocks/clusters.  So small files
actually waste space on disk in a "normal" file system.  Again, this is
something that more advanced file systems can be tuned to accommodate.

Maildir has constant time add and delete behavior - mbox is linear (mh

Hmm?  Shouldn't mbox's add time also be constant?  I mean, to add to the
file, doesn't "it" just jump to the end of the file and append the new

Yes, for a single thread of delivery - but when you have multiple
writers they line up behind a single mutex - hence, linear behavior.

directory - but neither one performs well in the large.  Maildir and

How large?  Really, I'd be interested to know this.  I'd consider 50 MB
pretty darn large for a mail file.

It's not.  In a past life I worked on the email infrastructure for an
online service with several million members.  Our test cases were
looking at sized ranging from a few hundred k to 100Meg.  Maildir starts
to choke (again, in the multiple delivery case) very early - around 300k
if memory serves (this work was done a number of years ago - the details
are getting fuzzy).  Maildir scaled nicely once we tuned the underlying
filesystem - we were able to push the bottleneck right down to the disc

mh can be tuned by "fixing" the filesystem.  Mbox can't be tuned.

Hmm, you mean Maildir & MH could theoretically be "fixed", right?  But I
see your point.

see above - there's nothing theoretical about it.  

Anyway - I think we've effectively beat the corpse into the ground - if
you've got more questions, feel free to send them to me off-line.

   Dan Berger [dberger ix netcom com]
   Nolite te bastardes carborundorum

   "We are what we repeatedly do.  Excellence, then, is not an act, 
    but a habit."
                                  -- Aristotle

    A982 E6B1 CB2F 7A49  843A 9297 DA73 4371  1F54 8D0C

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]