Re: New Server Hardware

Yes, you're right.

In fact it mostly depends on the ability of the filesystem/OS to 
efficiently list directories containing thousands of files, and to 
quickly read each of them to get relevant headers (Subject, To, From, 
Date...). On a lot of filesystems, this operation is very slow, compared 
to scan the same information within a single (indexed) mbox.

Since it's possible to combine multiple mailbox formats within a single 
account, it's possible to get the most of each format by using mslc for 
common-use mailboxes (inbox, sent items), and mdir for mailboxes 
connected to an external tool like you describe. Archive-type mailboxes 
could also use mdir to help reduce the size of incremental backups - a 
longer load time can be more acceptable for those mailboxes than it 
would for INBOX.


*Nicolas Hatier, ing.* <[email protected] 
<mailto:[email protected]>>
Niversoft idées logicielles - http://www.niversoft.com



On 2016-07-22 18:03, Bill Cole wrote:
> On 21 Jul 2016, at 18:06, Nicolas Hatier wrote:
>
>> mdir has the issue of not being indexed.
>
> In a sense, with mdir the filesystem IS the index. Each message is a 
> single RC822-format text file with its name encoding the message's 
> IMAP UID, multiple status flags, time of receipt (IMAP "INTERNALDATE") 
> and line count.
>
>> Opening a large mailbox with several thousands of messages may take a 
>> while, it all depends on the filesystem.
>
> Also on the ability of the OS to efficiently cache directory 
> information, which is a mix of how much memory pressure there is, 
> system tuning, and which filesystem is used.
>
>> All other operations are very fast. And it's very handy for 
>> incremental backups.
>
> Also, because each message is in its own file in standard RFC822 
> format with a name that carries significant metadata, mdir is 
> especially convenient for external tools. For example, I have a simple 
> script that feeds messages into the SpamAssassin learning subsystem as 
> spam when a user moves them into a folder with a particular name. 
> Without mdir that script would need to figure out how to extract 
> individual messages from mslc or mbox files and how to know which 
> messages are the new ones.
>