UNIX for Dummies Questions & Answers

I am not sure where to post this so i will put it in the newbie section.

I have set up a bog standard debain 6, LAMP environment in the cloud.

The specs

1 core at 2GH
2.5gb Memory

running Jommla, with about 1.6K visitors a day.

I am using AppFirst (appfirst.com) to monitor the server, and I am getting memory usage over 956M alerts almost every hour (not at the same time)

I have had a look at their monitoring tools and I can see the following.

1. At the times of the alerts the Page Faults go up very high.

2. The number of process goes up really high (i assume this is what is causing the page faults).

3. Most of the page faults are associated with apache.

I intialy thought it was the database , as joomla is bit database heavy sometimes.

However I am not sure now and I need some pointers.

I have no swap space defined on the disk, I totally forgot about this when setting it up.

Any kind of help would be good, the memory spikes may have nothing to do with the page faults but I think it does.

Since you have no swap, it is all from mmap()'d pages like libraries but can be any file with any app using mmap()/mmap64(). Page faults are not toxic, sometimes you can find an app that stirs things up unnecessarily and get rid of it, else buy more RAM.

You can make swap after the fact, by creating a big file and assigning it as swap. I would expect it is good for system stability to have some, if not lots. Some like to buy SSD for swap, but I am not sure as it is not churn friendly.

Mostly, page faults are when a process uses memory it hasn't needed before and the operating system goes "Hmmmm, I haven't actually given you that memory yet, hang on" and makes the process sleep a moment while it hunts up more memory for it.

It could also be happening when the system's low on memory and swapping, but since you don't have swap, that's not relevant.

It's possible to add swap even if you forgot to partition it, by creating a large, empty file, running mkswap on the file, and mounting the file as swap.

I'm still not sold on the idea of SSD's myself. There's a world of difference between "This mechanical disk will work for at least a year no matter how the software abuses it" and "every single write you do to your SSD slowly, inexorably kills it, but our algorithm spreads the damage enough it's unlikely to be problem."