I had something weird happen yesterday that will describe to see if anybody else recognizes this symptom. We have a Proxmox 2.1 server with a very low workload. There are two openvz containers running CentOS 5.x, and there is one KVM vm running Win2008. This is an 8 core machine with 16G of RAM, each container and vm is limited to 1 CPU and CPU units is unset, so defaults to 1000. The containers have always run nice and fast.
Yesterday morning, I got a call saying that the webserver running on one of the containers was running really slow. I verified that there was about a 13-18 second pause when connecting with a browser, then the page content would come down in about 4-6 seconds, bringing the total page load time to about 20 seconds on average. I straced the apache process and it was consistently taking 6 seconds to read in the php pages, then another 4 seconds to query the database and get the results, then another 2-4 seconds to generate the html, then finally another 4 seconds to feed it out. It was stupid slow. The host was pretty lightly loaded. Disk I/O was minimal. I couldn't find anything that would be causing this to occur.
Both of the containers were set to 1000 CPU units, and after throwing ideas around, I decided to set this particular container to 2000 CPU units to see if it helped any. Well, it sped right up to normal. To test, I set the CPU Units back down to 1000. It was _still_ fast (i.e. normal). Even now, the CPU utilization number shown above seems wrong. Both containers are set to 1000, so the one KVM (set to 1 CPU) is using the other 2000?
Question 1: Has anybody come across that before? Did something else possibly change that I'm not aware of when I changed that CPU Units section?
Question 2: The numbers for CPU units are ratios, right? so if I have 2 containers set at 1000 each (or unset as the case may be), that doesn't mean they only get 2000/1915120 of the CPU, instead that means they each get roughly half of the CPU. Correct?
...Todd
Yesterday morning, I got a call saying that the webserver running on one of the containers was running really slow. I verified that there was about a 13-18 second pause when connecting with a browser, then the page content would come down in about 4-6 seconds, bringing the total page load time to about 20 seconds on average. I straced the apache process and it was consistently taking 6 seconds to read in the php pages, then another 4 seconds to query the database and get the results, then another 2-4 seconds to generate the html, then finally another 4 seconds to feed it out. It was stupid slow. The host was pretty lightly loaded. Disk I/O was minimal. I couldn't find anything that would be causing this to occur.
Code:
ivcld51:~# vzcpucheck
Current CPU utilization: 4000
Power of the node: 1915120
Question 1: Has anybody come across that before? Did something else possibly change that I'm not aware of when I changed that CPU Units section?
Question 2: The numbers for CPU units are ratios, right? so if I have 2 containers set at 1000 each (or unset as the case may be), that doesn't mean they only get 2000/1915120 of the CPU, instead that means they each get roughly half of the CPU. Correct?
...Todd