Whoops! I failed to reply to the list. Sorry!
---------------------------- Original Message ---------------------------- Subject: Re: [Imap-uw] iowait when opening a mailbox? From: "Tom Cooper" <[EMAIL PROTECTED]> Date: Wed, August 17, 2005 12:06 To: "Mark Crispin" <[EMAIL PROTECTED]> -------------------------------------------------------------------------- Mark Crispin said: > The first thing that you should do is update your software. imap-2002d is an older version. The current release version is imap-2004e. I'll look into this. Apparently RH chose to switch to cyrus, which explains why there's not a substantially newer version of your software... > > In traditional UNIX mailbox format, it is necessary to read every byte of the file as part of startup. A 200MB mailbox is a lot to read. It may also be necessary to do some rewriting to assign UIDs and status to newly-delivered messages. I agree though that your 540MB disk cache should keep this from happening. That makes some sense to me. My INBOX folder is consistently empty or DARN close to it. (1-2 messages max) I do have SquirrelMail check all folders for new mail, so I guess an empty inbox is really not a help here... > > One step to do if you consistantly have a mailbox of that size is to switch to the mbx format. It greatly reduces the number of reads when the mailbox is opened, as well as the number of writes to update the mailbox. To get started on mbx format, read the following FAQ: > http://www.washington.edu/imap/IMAP-FAQs/index.html#4.5 Thanks! I'll look at that as well. One wrinkle is that I'm leveraging procmail, which doesn't require anything else to play nice with mbox, but I see that you folks have dmail, which may make the transition relatively painless. > > Next, you need to investigate if the I/O wait isn't due to something else. For example: > (1) imapd does a reverse DNS lookup on the client's IP address to get > its name. That should be ok. I tried that and found that I could easily reverse lookup the IP address/name of the internet-exposed NAT. (Between work and home they have a transparent proxy/NAT.) > (2) the c-client library attempts to launch IMAP-via-rsh to avoid the > need to negotiate authentication. Unfortunately, rsh tends to block for a minute if the rsh attempt is refused rather than quit immediately. Hmm. I'd expect this to fail 100% of the time. Any way to disable that? > If either of these are the problem, it will show up even with small mailboxes. > > Finally, does the problem happen with an IMAP client such as Pine? Even on my home LAN, Thunderbird seems to take it's sweet time to open the mailbox. I don't think it's as bad as the webmail over https, but it can be slow. I have not investigated whether this is related to local resources or load on my server. I can give it a shot with Pine. Unfortunately if I run pine, it will most easily be from the mail server itself, which would eliminate the reverse-lookup issue - wouldn't it? I've got a couple of other linux boxes at home. I can configure pine there to see what happens.... > I > don't know whether squirrelmail does this, but many webmail programs literally make a separate IMAP connection on a per-click basis. This puts a tremendous I/O load on the IMAP server. Only a few very well-written webmail programs maintain a static IMAP connection in webmail. I'll inquire with the squirrelmail team to see if there's an option for that. Thanks much for the great feedback! Regards, Tom Cooper _______________________________________________ Imap-uw mailing list Imap-uw@u.washington.edu https://mailman1.u.washington.edu/mailman/listinfo/imap-uw
