Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

This question already has an answer here:

I run an Ubuntu 8.04 on a Slicehost virtual server with some lightweight server apps - apache22, svnserve, mysql, and proftpd. The only serious service limitation is RAM - 256MB is what I'm paying for.

I noticed that if I let the system run for a few days/weeks, the amount of free RAM slowly declines, and paging file is being used soon after. For example, upon rebooting I may have 60% of RAM free, next day it may be at 55%, etc.

             total       used       free     shared    buffers     cached
Mem:           256        114        141          0          3         50
-/+ buffers/cache:         61        194
Swap:          511          0        511

How would I prevent the amount of available memory to decline?

Edit: Here's my ps -aux listing the top memory consumers. I left out all the system processes. I can see that apache and mysql top the memory usage.

    USER       PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
    root      1369  0.0  0.3  16844   952 ?        S<s  00:10   0:00 /sbin/udevd --daemon
    syslog    2110  0.0  0.2  12288   748 ?        Ss   00:10   0:00 /sbin/syslogd -u syslog
    root      2131  0.0  0.2   8128   588 ?        S    00:10   0:00 /bin/dd bs 1 if /proc/kmsg of /var/run/klogd/kmsg
    klog      2133  0.0  0.4   4516  1304 ?        Ss   00:10   0:00 /sbin/klogd -P /var/run/klogd/kmsg
    root      2154  0.0  0.4  50904  1152 ?        Ss   00:10   0:00 /usr/sbin/sshd
    root      2211  0.0  0.2   3932   592 ?        S    00:10   0:00 /bin/sh /usr/bin/mysqld_safe
    mysql     2253  0.0  8.8 161940 23252 ?        Sl   00:10   0:00 /usr/sbin/mysqld --basedir=/usr --datadir=/var/lib/mysql --user=mysql --pid-file=/var/run/mys
    root      2254  0.0  0.2   3836   604 ?        S    00:10   0:00 logger -p daemon.err -t mysqld_safe -i -t mysqld
    proftpd   2340  0.0  0.5  63000  1504 ?        Ss   00:10   0:00 proftpd: (accepting connections)
    root      2356  0.0  0.3  18608   964 ?        Ss   00:10   0:00 /usr/sbin/cron
    root      2384  0.0  0.3  83360   892 ?        Ss   00:10   0:00 svnserve -d -r /etc/svn/svn-rep-01
    root      2387  0.0  3.5 194628  9220 ?        Ss   00:10   0:00 /usr/sbin/apache2 -k start
    root      2409  0.0  0.2   3852   576 tty1     Ss+  00:10   0:00 /sbin/getty 38400 tty1
    root      2410  0.0  0.5  15252  1444 ?        Sl   00:10   0:00 /usr/lib/ruby/gems/1.8/gems/passenger-2.0.2/ext/apache2/ApplicationPoolServerExecutable 0 /us
    www-data  2411  0.0  3.4 195880  9100 ?        S    00:10   0:00 /usr/sbin/apache2 -k start
    www-data  2412  0.0  7.3 205532 19400 ?        S    00:10   0:00 /usr/sbin/apache2 -k start
    www-data  2413  0.0  3.3 195620  8824 ?        S    00:10   0:00 /usr/sbin/apache2 -k start
    www-data  2414  0.0  3.4 195880  9080 ?        S    00:10   0:00 /usr/sbin/apache2 -k start
    www-data  2415  0.0  3.4 195888  9056 ?        S    00:10   0:00 /usr/sbin/apache2 -k start
    root      2416  0.0  2.4  43448  6512 ?        Sl   00:10   0:00 Passenger spawn server
    www-data  2437  0.0  7.7 208116 20248 ?        S    00:48   0:00 /usr/sbin/apache2 -k start
    www-data  2519  0.0  3.3 195644  8820 ?        S    02:12   0:00 /usr/sbin/apache2 -k start
    root      3026  0.0  1.1  67960  2892 ?        Ss   20:53   0:00 sshd: dv [priv]
    dv        3028  0.0  0.6  67960  1700 ?        S    20:53   0:00 sshd: dv@pts/0
    dv        3029  0.1  0.8  19392  2304 pts/0    Ss   20:53   0:00 -bash
    dv        3041  0.0  0.4  15056  1092 pts/0    R+   20:54   0:00 ps -aux

Overall, I'd like to thank everyone for their thoughtful answers and it was really hard to choose the best because each has some useful information in it.

I'll look into moving to lighthttpd or nginx, or at least reduce apache's MaxClients parameter.

share|improve this question

migrated from stackoverflow.com Oct 26 '09 at 3:21

This question came from our site for professional and enthusiast programmers.

marked as duplicate by Michael Hampton Jul 31 at 19:56

This question has been asked before and already has an answer. If those answers do not fully address your question, please ask a new question.

    
Follow-up: I am now running nginx + php + php-fpm + xcache, total memory usage with conservative settings is < 64 MB. –  DV Mar 10 '09 at 4:46
    
"Why is my computer using memory rather than wasting it?" –  David Schwartz Jan 15 '13 at 17:04

9 Answers 9

You will notice that some of that is being consumed by "cache".

This is just disk read/write being cached and you can basically assume its free because it will be dumped out as soon as something more important needs it.

It makes disk IO faster. :)

I have presently 2G of disk cache on my box, makes things fast.

"free" is a poor memory analysis tool. I find "htop" a much more practical overall system management tool, which gives both a practical representation of memory use and a practical way to explore processes.

Like this, just prettier:

1  [|||||                                                                          5.0%]  Tasks: 156 total, 2 running
2  [|||                                                                            2.5%]     Load average: 0.29 0.30 0.31
Mem[|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||     1182/3967MB]     Uptime: 5 days, 16:25:36
Swp[||                                                                        37/2000MB]
Mem:3967M used:1182M buffers:0M cache:1952M

( if you're wondering about the low uptime.. thats a laptop )

A lot of people might provide ways to negate caching to get stuff out of swap, but its more or less pointless. All you'll end up doing is reducing performance, unless you have a program that is really memory intensive , but only runs intensively every few days so that when its not running intensively it gets swapped out, and when it does come on, you cant wait the ~.5 seconds it takes to swap it back into memory from disk.

I personally can't think of any good program that matches these conditions. shrugs

The people whom wrote the kernel know more or less what they're doing in this department, I'd trust their judgment unless you're certain you know better.

as @Paul Betts said, Switching to lighttpd might help you save a bit. Depends on what you're doing, there are always a few tradeoffs.

I'm using it on my vps. Only has 96M ram and 63M swap, and its only using 26M(no joke) and 35M respectively. And there's a database running there as well. ( Postgres , but not really used )

share|improve this answer
    
Is there something similar to htop on a OS X? I use top. –  jpsilvashy Oct 7 '09 at 21:56
1  
@jpsilvashy: something like geektechnica.com/2009/07/how-to-install-htop-on-osx ? –  Steve Schnepp Nov 11 '09 at 9:34

Ignore the first line. The important numbers are on the "-/+ buffers/cache" line. Linux will clear cache and buffers to make way for processes.

My Debian server routinely shows between 2 and 10 megabytes of free space on the first line. The second line shows that I have 50 megabytes free. (With a total of 256 ram on the system)

Your numbers look fine. DO NOT clear your buffers/cache. You have more than enough free ram.

Now if you have no memory free for cache, your system will run slow. If you start using alot of swap you're going to have problems serving webpages.

share|improve this answer

you need to change your swappiness (tells the kernel how much it should swap)

# sysctl -w vm.swappiness=0

setting it to 0 will tell the VM to keep as much data in memory as possible, but you need to go and read up on whats right for you, its hard to tell, but do some experimenting and see what you can come up with.

edit the vm.swappiness line in /etc/sysctl.conf to set it up at boot

in other words, you want to use the memory as much as possible and reduce the amount of swapping that occurs on disk

share|improve this answer
    
This will reduce the amount of swapping, but it will increase the amount of paging. (More dirty pages in the same amount of memory will mean fewer clean pages in memory.) On balance, it will make things worse. Please don't do this. –  David Schwartz Jan 15 '13 at 17:03

Use lighttpd instead of apache - this doesn't solve your memory leak, but will help you save RAM. Also, you should use top/htop to figure out which process is doing the leaking, then you can debug it further. There are 1000 reasons why this could happen - you need to investigate.

share|improve this answer

If it always cuts into the swap until that is full, you probably have a problem. As long as it's just light use in the swap, and the physical memory is full, that's ok. You want all your physical ram used up, that's what it is there for.

share|improve this answer

The answer depends on what's eating your RAM. Start narrowing down the problem by comparing the output of two ps aux runs 24 hours apart.

Also, be aware that Linux uses memory heavily to cache I/O. Those caches are freed as soon as another process asks for that space, so they shouldn't ever negatively affect performance. That means the second line in free's output (-/+ buffers/cache) is the important one.

share|improve this answer

What's your MaxClients setting in your Apache config? My Slicehost VM did the same thing until I dropped mine down to something sane for the amount of RAM I have on the box.

share|improve this answer

It's a bit of a plug but hopefully a good one to save on resources ;)

Nginx (better than Apache) Web Server - VPS BIBLE Pt 11

.. this is tested on Linode which is a similar setup to Slicehost .. but they do provide 360MB RAM for the same $20 a month (and an extra 6 gB space), which is what you'll be paying, DV, for your plan.

Another part of teh VPS Bible, linked in the full series index on the page, sets out how to configure Xcache, if that helps.

Otherwise, you can play with swappiness without having to reboot, pretty key for a server:-

sudo sysctl vm.swappiness=10

Just change "10" for whatever value, depending on your needs. Reenter the command with different values. When eventually you reboot, value defaults.

share|improve this answer

You can use:

echo 1 > /proc/sys/vm/drop_caches

I believe to free the caches as well. It should drop the caches entirely, but after a few days you'll be right back where you are now. You could cron it up though.

share|improve this answer
    
You could, but please don't. That will cause huge amounts of memory to be wasted. –  David Schwartz Jan 15 '13 at 17:03