RAM utilization issues


#1

Hi,

Throughout the last few months we’ve been slowly migrating off our old VMware installation to KVM, most of it has been successful and people has been happy with the end-result.

The new infrastructure obviously has a bit more of power which is heavily utilized by the guests however compared to VMware the RAM usage is way above what we’ve expected.

In our VMware environment we operated at 24GB with ballooning and we were able to maintain usage at around 90% without any swapping. However with the new environment we’ve 24x3 GB och RAM and we’re seeing a 80% utilization across the board with the same amount of guests.

Please notice that we’ve both ballooning (on both the host and guest) and KSM enabled, hugepages hasn’t made any major difference so we’ve chosen to disable this. If you’ve any input as to why we’re seeing this behavior or if it’s expected we’d appreciate it. In addition to this we’re utilizing virtio_blk/net/pci/ring and vhost_net.

Some data:
root@kvm01-01:~# grep -i balloon /boot/config-$(uname -r)
CONFIG_VMWARE_BALLOON=m
CONFIG_XEN_BALLOON=y
CONFIG_VIRTIO_BALLOON=y

Here’s our KSM history from last week on the servers:
oi43.tinypic.com/2s9d56v.jpg
oi41.tinypic.com/ru8lm9.jpg
oi43.tinypic.com/2s9d56v.jpgHi,

Throughout the last few months we’ve been slowly migrating off our old VMware installation to KVM, most of it has been successful and people has been happy with the end-result.

The new infrastructure obviously has a bit more of power which is heavily utilized by the guests however compared to VMware the RAM usage is way above what we’ve expected.

In our VMware environment we operated at 24GB with ballooning and we were able to maintain usage at around 90% without any swapping. However with the new environment we’ve 24x3 GB och RAM and we’re seeing a 80% utilization across the board with the same amount of guests.

Please notice that we’ve both ballooning (on both the host and guest) and KSM enabled, hugepages hasn’t made any major difference so we’ve chosen to disable this. If you’ve any input as to why we’re seeing this behavior or if it’s expected we’d appreciate it. In addition to this we’re utilizing virtio_blk/net/pci/ring and vhost_net.

Some data:
root@kvm01-01:~# grep -i balloon /boot/config-$(uname -r)
CONFIG_VMWARE_BALLOON=m
CONFIG_XEN_BALLOON=y
CONFIG_VIRTIO_BALLOON=y

Here’s our KSM history from last week on the servers:
oi43.tinypic.com/2s9d56v.jpg
oi41.tinypic.com/ru8lm9.jpg
oi43.tinypic.com/2s9d56v.jpg