| High Load Issues [message #12615] |
Fri, 04 May 2007 11:56  |
aeterna
Messages: 3
Registered: May 2007
|
Junior Member |
|
|
I have been experiencing high load on my HN. Sometimes the load get's as high as 100+. I've narrowed it down to inproper configuration files for our VE's however I was hoping someone could just confirm my suspicions. I just started reading up on everything the wiki has but its still alot of information that im catching up on..
Here is an example of what one of our VE's UBC parameters look like:
# UBC parameters (in form of barrier:limit)
# Primary parameters
AVNUMPROC="80"
NUMPROC="385"
NUMTCPSOCK="160"
NUMOTHERSOCK="2147483647"
VMGUARPAGES="12288:2147483647"
# Secondary parameters
KMEMSIZE="16777216:16777216"
TCPSNDBUF="4880000:8000000"
TCPRCVBUF="4880000:8000000"
OTHERSOCKBUF="132096:336896"
DGRAMRCVBUF="1058384:1058384"
OOMGUARPAGES="16144:2147483647"
# Auxiliary parameters
LOCKEDPAGES="32:32"
SHMPAGES="8192"
PRIVVMPAGES="351072:422978"
NUMFILE="14280"
NUMFLOCK="100:110"
NUMPTY="36"
NUMSIGINFO="256"
DCACHESIZE="5916672:6163200"
PHYSPAGES="0:2147483647"
NUMIPTENT="2147483647"
So if i take 12288*4096 = 50331648 bytes = 48 MB = This is my guaranteed memory limit?
while my burstable is something like 1.3GB, correct?
[Updated on: Fri, 04 May 2007 11:57] Report message to a moderator |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| Re: High Load Issues [message #12660 is a reply to message #12658] |
Sun, 06 May 2007 18:07  |
rickb
Messages: 368
Registered: October 2006
|
Senior Member |
|
|
I think its worth noting--
As Dev pointed out, you do not have any cpulimits set on your VEs. This is generally bad especially if each VE has its own admin; ie you want to ensure fariness rather then optimal overall efficiency. When you add cpulimits to your VEs, the load will not decrease, and in fact it may increase. This is because a few of your VEs will reach their cpulimit and their runqueues will expand, which means more processes want to run as seen by your HN.
However, the end result is your cpuunits will kick into gear and each environment will have its amount of cpuunits cputime/equivalent MHz to count on. Each of your VEs will feel more interactive when cpuunits are well enforced, and cpulimit will protect the HN from reaching 0% idle time very often, which makes things even more interactive.
Be sure that your cpuunits are not oversubscribed. See my message #8620 on this page to read more about cpuunits/cpulimit and how they relate to each other. http://forum.openvz.org/index.php?&t=msg&th=1551
The bottom line is that the load as seen by the HN is not always an accurate factor in determining if your server is overcommitted, underpowered, etc (you may have plenty of idle cpu time with a load of 100!). With regard to strictly your application load, it may and probably is overcommited; however if you are not the admin of your VEs, like in the case you are selling them as a service, you are fulfilling your responsibility in guaranteeing them a time slice of the cpu / equivalent CPU speed in MHz and the HN load is not something you can control.
Hope this isn't too wordy!
-------------
Common Terms I post with: http://wiki.openvz.org/Category:Definitions
UBC. Learn it, love it, live it: http://wiki.openvz.org/Proc/user_beancounters
|
|
|
|
OpenVZ Forum
Members
Search
Help
Register
Login
Home
plz do the following:
