I'm trying to configure a collaboration suit (PHP, PostgreSQL, Apache) plus a mail server(Sendmail and Dovecot) on a single box. It will be used heavily 24*7. So having a long up time is really needed. I'm looking for 3 to 6 month up time. I want to know about kernel update frequency(because they are causing server reboot) for different operating systems such as Linux(CentOS, Debian, Ubuntu Server Edition), FreeBSD NetBSD, OpenBSD.
Could anyone help? I'm not trying to make a Linux vs BSD question as it is against forum rules.
Disclaimer: I can only attest for the Linux side of things here.
First of all, 3 to 6 months isn't that long of an uptime, as I've seen systems with 2+ years. And with the distributions you've listed there won't be any major kernel updates in that time, save important security updates, and even those don't force a reboot (but the new kernel will only be used after a reboot). If you really need high availability, better to look into proper redundancy by setting up a cluster, where a second node can easily take over should the primary suffer a hardware defect.
Unfortunately I'm not a cluster guy :D. On my home server I use CentOS but i see that there is a kernel update each month. Should I reboot each time? Where could I find information about each update. Is it a security or a normal and I can postpone reboot.
As always: it depends. Information about security updates is mostly relayed through special mailinglists, and CentOS and Debian (to my knowledge) only release updates for security problems or severe bugs for any release. Meaning, any software installed for CentOS will stay at that version and only receive security fixes, from the kernel to the GUI.
Whether or not you can postpone a reboot depends on how risked you estimate your system to be.
It is entirely depend on you and your applications criticality and compatibility, if you want to have new kernel every time. In my environment, it is not so much required. If you can afford little downtime, then Solaris Live Upgrade is a feature in Solaris, where you will update your kernel without having long downtime. It will be up with just a reboot after setting up all required things. I am not sure, if you have Solaris in your option.
First of all, I quite agree with the cluster idea.
If it is really of the question, put whatever distro you want and simple DON'T update kernel. You are not forced to apply each update of the distro. Actually in a production machine, you don't update at all, at least not automatically. If for example the "automatic" update request new version of apache, your machine will have a small down time. If problems with the new configuration files arise, then it will be big down time.