[Tfug] An Ext4 question...

[Choprboy]

On Sunday 09 August 2009 15:39, Jim March wrote:
> > How big of mailboxes are you talking about?  (disk size/# of messages)
>
> We're talking inbox falling apart at 2gig.  It's a lawyer friend of mine.
>


Being a lawyer (and this would go for any business email box I suppose)... I 
would assume a large number of the emails have attachments, long Re: 
sections, or otherwise exceed 4K each. But I would expect the "falling apart 
at 2G" has to do with the way MBox works, not the individual message size. 
Since each MBox mail folder is a single file, once you hit 2G the app 
processing the file may have a 32bit memory access problem (i.e the 
programmer used "int" instead of "long int" or "huge int" for their file 
access pointer).

Personally, I think the reason to go to MDir over MBox is access time. With 
MBox, the entire folder file must be parsed on every access. Unless you have 
a relatively small folder, it will suck up huge amounts of CPU.

My own mailbox is on MDir (years ago I used MBox, but I had to abandon it as 
my folders grew). I currently sit at a roughly a mere ~88,500 email messages, 
my largest folder being about 56,000. I would guess more than half are less 
than 4K in size, with an apparent size of 515M (all the actual file sizes 
added up). True on-disk space used is 670M.

Probably the biggest problem is that my indexes have now grown to 27M on my 
largest folder, the initial opening takes several seconds (though once cached 
it's fast). The next step up would be something like Cyrus, though that is 
server based, not local.


Adrian