OpenVZ Forum


Home » General » Support » But why is the RAM gone?!
Re: But why is the RAM gone?! [message #26935 is a reply to message #26917] Tue, 05 February 2008 15:52 Go to previous messageGo to previous message
HubertD is currently offline  HubertD
Messages: 22
Registered: August 2006
Junior Member
If the memory would be used for caching, then it should appear somewhere in "Buffers" or "Cached" in /proc/meminfo, I suppose?

The only thing that I'm really sure of is the following situation:

- After a reboot, free memory (including buffers and cache) reported by "free -m" starts at ~2GB
- This value is constantly decreasing over time, until there is almost none left after 6-7 days or so
- At this point, the server's load goes up dramatically (load>50), I suppose because of swapping, but I don't have prove for that.
- "Monit" on the server sees the high load and reboots the machine.

Before monit was installed, the machine became unresponsive and I wasn't able to log in via SSH or serial console any more.

I observed this cycle for about 10 times now.

During the cycle, the sum over "RSS size" of all processes (as returned by "ps aux") wouldn't increase, in fact, it decreases.

I'm almost at the end of this cycle (and my nerves ;) again and, did a snapshot of /proc/meminfo, /proc/user_beancounters, "ps -aux" etc. this noon.

I combined the data using openoffice calc, found out nothing new but at least it looks a little bit clearer as a spreadsheet:

http://www.denkmair.de/ramgone/ubc.png
(please excuse the picture, but you wouldn't like the super-messy html-export either ;)
The full document is available here: http://www.denkmair.de/ramgone/ubc.ods

If I understand the UBCs right, the sum(kmemsize+buffers+privvmpages) should reflect the total RAM usage on the machine, caches excluded.

So UBC tell me that I got ~2GB memory used.
"ps aux" sums up the process "resident sizes" to only about 1.4GB, which may go along with the 2GB from UBC.

But /proc/meminfo & "free" tell me that >3GB are actually used, buffers and caches not included.
THAT and the unresponsive machine when the free memory reported by /proc/meminfo reaches zero makes me sure that there REALLY IS a problem ;-)



 
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Read Message
Previous Topic: Shutdown of VE affects interface on HW-node?
Next Topic: Starting & Stopping Multiple VPS...
Goto Forum:
  


Current Time: Sat Jul 13 11:13:28 GMT 2024

Total time taken to generate the page: 0.02236 seconds