I'm working on a uni exercise which I then should implement using virtual machines.

Some aspects:
1. A company wants to implement an e-mail service.
2. The company has several sites.
3. Employees should be able to access their e-mail from any of the sites.
4. If one site is destroyed, e-mails should still be available from other sites.
5. Users will access their e-mail by making requests to dedicated e-mail servers using the IMAP protocol.

My first approach would be to install dovecot on the e-mail servers, and have dovecot access a distributed file system such as GlusterFS, OpenAFS, NFS...

My problem is that the above mentioned file systems offer many pros and cons, and I am ignorant as to what standard solutions are currently being deployed in real organizations.

Is this approach correct? Which distributed FS would be optimal for this situation?

Smartass answer - Outsource it, everyone does these days! Google or Office365 FTW!!!

And more realistically, what research have you done so far and what have you tried setting up?

First thing to realize is that "optimal" cannot be quantified. What is best for ONE company/situation/environment, may not work AT ALL for another. All of this depends on the hardware you're using, how you're connecting to the other disks, etc.

For example, if you have your dovecot server running on a bunch of blade servers, all sharing a backplane with huge bandwidth, connected to a SAN disk, what is 'optimal' for that is different than several separate servers, with only a single NIC. Your questions 1-5 allude to "sites", but don't say if the email services are in the cloud, hosted at ONE site but AVAILABLE to all, or if there's an email server at EACH site that you want to act as a cluster. ALL of these things come into play.

What happens in 'real organizations', is that such things are put in data centers, and things are accessed from there via dedicated WAN connections. These will typically have disaster-recovery sites, in case the primary data center goes down.

Another thing to consider is that a distributed File System isn't necessarily the way to go here, I can think of at least one other approach.

1 members found this post helpful.

I hear fail-safe and fool-proof, but saw "distributed"

"So can I," (he said, cryptically).

Since this is a University (homework ...) exercise, why don't you first sit down and come up with something and – after checking with your teacher to be sure it isn't "cheating" – bounce it off of us. "Do your own leg-work first."

Do research, yourself, and then interpret your findings, again, yourself.

This is a very realistic "real world" scenario and it's therefore important that you work out the issues that are set forth in the exercise's several bullet points, "thinking outside the box" as you do so. (The exercise apparently did not stipulate "a network file system," now did it? I believe that you assumed that part ...)

This is something that you very easily could be called upon to do, or to be a part of, "on the job." And, on the job, you'd be doing research. You'd also be looking for every technical option. You could easily "jump to a conclusion" and never think to re-consider it. (Lots of faulty implementations are made that way, and they get deployed!)

- - - - -

Your instructor has crafted a very well-designed exercise and it is very important that you go through the steps that s/he has in mind. Talk with him/her for guidance, and take your initial musings to him/her before you bounce them off of us or any others. Always keep your instructor "in the loop." (S/He needs to know what the students are thinking, in order to guide them to successful learning.) "I know: I am one ..."

Don't short-cut to "the right answer." This exercise is leading you through a very relevant and business-important discovery and decision process that you will go through (in various forms) many, many times "on the job." Make the very most of it. You'll thank me.

1 members found this post helpful.