Openstack (Queens) instance (VM) running deadly slow [closed]
Folks, in a four nodes Openstack (Queens) installation, I am experiencing instances running deadly slowly - It does not matter the size of VM, from 1-8vCPUs and 2-16GB of RAM. Indeed, as bigger the instance the slower is. I am using Sysbench as a reference and tried with Ubuntu 18.04 and CentOS 8.1. Just to have an idea, on nodes with Westmere Intel CPUs with a VM of the same size, one running VMWare ESXi and the other Openstack (KVM/QEMU), the VMWare node runs 10+ times faster and 10+ times lower latency than KVM/QEMU.
Initially, I found that my VMs were identifying the wrong CPU architecture, executing "lscpu" inside the VM showed an AMD EPIC architecture instead of Intel Westmere. However, I already changed the nova.conf cpu_mode=custom and specified cpu_model=Westmere. Now each VM is reporting correctly the CPU architecture.
Operations like "yum update" is taking almost two minutes to start showing up something at the screen and shutting the guest CPU to 100% all time.
Any ideas of what is the root cause of this problem? where should I start looking? debug strategies?
Best Regards! -Bruno Domingues
A few suggestions. Check: Is the CPU overloaded? Memory? Disk? Or is it a networking problem? Check the relevant figures on the VM and the host. Run a CPU-bound program and compare its performance with the same program on an ESX VM. Can you identify classes of programs that are slow?
I found the problem. I was using qemu as hypervisor instead of kvm. As soon I changed the configuration into nova, rebooted all nodes and defined host-passthrough into nova config file, it worked like a charm... as fast as my VMWare guest.
Thumbs up! Good work.