On 3.12.2011, at 22.30, Maria Arrea wrote:

We are using dovecot 2.0.16 with mdbox+zlib. We are now testing SiS (Single Instance Storage) and I have 5 questions:

1º Is possible to "dedup" existing mdboxes?

You can dsync the mailbox elsewhere and then replace the original with the new copy.

2º Are attachments compressed with zlib if mdboxes already use zlib?

Currently attachments don't support zlib at all.

3º I have plenty CPU to spare, should I use a low value of mail_attachment_min_size , like 16KB ?

It wastes disk seeks since it now has to read mail from 2 (or more) places in filesystem, so probably not a good idea. So in any case SiS most likely increases your disk IOPS usage.

4º Can I "undo" SiS if I have problems?

dsync will help the other way around too.

5º We use bacula to save indexes & mdboxes, and we recover mailboxes using doveadm import when a user "makes a fatal mistake" wiping all her Inbox. If we enable SiS I am not really sure how can we safely restore a user's INBOX if that user has SiS attachments.

Hm. Yes, that is problematic.. Even if you knew what SIS files were used, there's no simple way to restore those with proper refcounts. I think what really should be done is writing a tool that can create/restore backups, possibly de-SISing the attachments..