Hi list, I've been perusing the list archives, but I couldn't find a recent, definitive answer to what I'd like to know. I'm looking to set up a new batch of servers to migrate ~160,000 accounts, spread across several domains, onto (from CommuniGate). Last year I worked with the earlier beta/RC versions of Dovecot, so I'm generally familiar with the setup required. However, a lot of development has been done since I was last dealing with it. So, what I'd like to end up with is a load balanced group of servers providing pop and imap access with quota support using an NFS filesystem. We have a new Netapp device (forget the actual model) that can do SAN/iSCSI/NFS. We tried using various clustered filesystems previously, but the performance/stability issues were a nightmare, so I've fallen back to NFS (or perhaps SAN if users can't hit any server in the group). So, what does my dovecot setup need to be? I've seen some people say they disable indexing in the LDA, but allow pop/imap to index. Some say the attribute cache needs to be disabled, which I'd think would be worse than just disabling indexes. Some say indexes should be kept on local disk (dedicated server for users), and others seem to report problems even when doing so. Is there a general consensus? I was also wondering if dovecot's sanity checking on the indexes would allow each server in a balanced group to keep indexes on local disk (so if the maildir has changed on another server, dovecot rebuilds it). If so, does that take away all performance gains from having indexes at all? Thanks for any info you all can provide.
On Tue, 2007-01-23 at 09:29 -0500, Justin McAleer wrote:> > > So, what does my dovecot setup need to be? I've seen some people say > they disable indexing in the LDA, but allow pop/imap to index. Some > say > the attribute cache needs to be disabled, which I'd think would be > worse > than just disabling indexes. Some say indexes should be kept on local > disk (dedicated server for users), and others seem to report problems > even when doing so. Is there a general consensus?I don't think there are any problems with keeping indexes in local disk, other than what generic index file bugs there may be.> I was also wondering if dovecot's sanity checking on the indexes > would > allow each server in a balanced group to keep indexes on local disk > (so > if the maildir has changed on another server, dovecot rebuilds it).Yes, it would.> If so, does that take away all performance gains from having indexes > at all?If the server changes often, probably.. I think your possibilities are (best performance -> lowest performance): a) Try to redirect the users into the same server always, so that you can keep indexes in local disks. b) Try to redirect the users into the same server, but keep index files in NFS. You don't need to disable attribute cache, as long as you're not redirecting the same user into two different servers at the same time (or within NFS attribute cache TTL more exactly). c) Disable index files and redirect the users wherever. -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 189 bytes Desc: This is a digitally signed message part URL: <http://dovecot.org/pipermail/dovecot/attachments/20070123/787a2ba4/attachment.bin>
Apparently Analagous Threads
- keeping indexes in tmpfs
- mulit user system redirecting My Documents
- Centos 7 kworker uses 100% of single core on mulit-core processor usage inquiry
- Centos 7 kworker uses 100% of single core on mulit-core processor usage inquiry
- Centos 7 kworker uses 100% of single core on mulit-core processor usage inquiry