OpenVZ can limit cpu usage, probably that's why you have a load of 58 (on the main node I guess). It's a normal thing to see the load average on a node higher when a VPS has a higher load average. Check how load avg is calculated in Linux.
If the VPS is doing IO (it's not clear from your details) the node will be affected and vice-versa. Other virtual servers are affected too since there is no I/O isolation (forget about ioprio, it's doing nothing in 99.9% of cases).
█ IntoDNS - Check your DNS health and configuration
█ FLEIO - OpenStack billing and control panel for service providers
Thanks Cristi4n, you gave me a light! Do you mean after I limit cpu usage for a vps, it is normal to see the load on main node can be high?
The I/O result is on main node. Is there any way to make sure which vps or process is causing I/O problem? Then I can suspend it manually at least. Thanks!!
Yes, it's normal -- what's happening is you've limited a VPS to, say, 1 core on a 4-core CPU, and that 1 core is backed up with all of the VPS's processes, while the other 3 are comparatively idle. Thus the total load average will be offset by the 1 backed up core.
To check which container is causing a problem, ssh to your node as root and do:
vzlist -o vpsid,laverage
This will output load averages for each VM, you should then login to each suspect (the most active containers) and check top/iowait (there may be an easier way to do this).
Don't use dsync for your tests as it's a pretty bad test of performance on SATA disks.
you should be using fdatasync if anything
BuyVM - OpenVZ & KVM Based VPS Servers - Chat with us
- All popular VPN methods supported
- Affordable offloaded MySQL & DDoS protection
- 5GB backup space, unmetered private LAN bandwidth & native IPv6 included. All with a strong serving of pony