So I’ve started getting some users with large (~2GB) IMAP accounts. At various times throughout the day the load creeps up from it’s normal 1 to about 5-10 based mostly on the IOwait usage.
I’m not sure if there is a way to make IMAP not put a D priority when reading the directory so that it doesn’t take over the server. Or some other config I can do to lessen the load?
If that isn’t possible, how hard do you think it would be to setup InterWorx to use /home for website and a /email partition on a separate SSD drive?
I’m guessing I would have to do some hard links manually after setting up an account?
Or I guess I can just tell my users archive emails and not allow inboxes that large.
Not an easy answer I don’t think to your post
If it were our systems, I might consider clustering, and set email to node (but this uses the same /home as CM)
Ssd may not work as fast as you think, which j think your on 15K drives, and the constant read/writes would be high I think, so lowering life of ssd
I would be surprised if moving mail folder onto its own hard drive would be easy, even with symlink
Probably best to have clients download their email onto PC and then delete email, to lower size of mailbox
I believe Microsoft restrict mailbox size on exchange, but these are dedicated mail platforms anyway
If I think of anything more, I’ll post but good luck
Thanks for the feedback!
Yeah, I figured the move wouldn’t be easy within the InterWorx setup.
I have 7,200K drives. I could maybe see the price for 15k, since that should be cheaper than SSD and maybe just make those my main drives for the whole server.
I think SSD would still be faster in my example, but if 15k can solve the issue that’s all I care about.
Ah sorry 7.5K drives, understand
If your looking at ssd, then 1gb are expensive (enterprise class) and there are circumstances which you are better to use plater drives (15K)
We have introduced a new server, using hybrid drives (ssd + plater - 7.5K), which are working very well, using a 6 drive raid array in raid 1, each drive is 1Gb, so that gives us 3 x 1gb drives
This appears to be working very well (touch wood), and largest spike we have seen is just under 2 (about 1,8), so you may want to consider similar
If it helps you decide, we can let you test on a old server we are bringing off line, and you can load on a 2gb mailbox and see how it performs and we can load a test server using ssd3, 250gb as well if you want
Thanks for the offer John, I just told the few users (all under one account/domain) to archive their inboxes or we would have to move them to an Exchange or similar hosted system. When I’m charging shared hosting prices, I have to set some limits somewhere.
My first thought is that SpamAssassin or any other spam defense is the main cause of load spikes, especially if you’re using Bayes on a site-by-site basis. But I’m sure there’s more to your case than meets my eye. First thing I’d do is what you already did and set rules about house cleaning.