This is kind-of a follow on from http://communities.vmware.com/thread/52352
But
Is memory Ballon (on hosts) or Swap (on Hosts) always bad news?
I have a 3-host cluster, and if I switch on DPM for 1 host, vCenter switches off that one host (presumably as all the running VMs can be accomodated on the other 2 hosts).
When all (50-ish) VMs are running on the 2 hosts, the hosts are up at around 90% Memory utilisation. Memory balloned per hosts increases steadily.
I'm going to run through the Resource Management guide late, but I just wondered if anyone had any thoughts on this?
My own thought was that if DPM powers-off one host, then the VMs are happily running on the 2 hosts, even with the Memory Ballooning?
Ian
Is memory Ballon (on hosts) or Swap (on Hosts) always bad news?
Ballooning is not necessary bad but it is an idication that things could go bad quickly ballooning is a sign that host is running out of memory - if your host is swapping that is bad it will affect the performance of the VMs
My own thought was that if DPM powers-off one host, then the VMs are happily running on the 2 hosts, even with the Memory Ballooning?
That is correct - the two hosts are fine and are using ballooning to free up memory -
If you find this or any other answer useful please consider awarding points by marking the answer correct or helpful
Thanks - yeah thats what I thought. If a host has memory 'Swapping' then it's going to local Disk (these hosts are ESXi embedded on SD cards).
What I found a little confusing was one host was showing around 9GB Balloon (on the perf graphs) and the other was 5GB. Each host has 32GB of Physical RAM, so around 30% of the host RAM was showing as Ballooned.
I would imagine that the amount by which you can 'overcommit' the memory depends on the nature of the VMs you have - for example if you mostly have Server 2003 VMs, they will have more memory pages in common which can be shared?
That is correct - but MS reuses code enough you will gain enough
efficiencies even between different version of windows
Sent from my iPhone
On Mar 29, 2010, at 8:14 AM, 1an3 <communities-emailer@vmware.com
Is the 9GB of ballooned memory on that one host a problem?
It shouldn"t be - the potential impact will really be on the vms it is
happening to since they are swapping to their own o/s swap file
Sent from my iPhone
On Mar 29, 2010, at 9:33 AM, 1an3 <communities-emailer@vmware.com
Do you have hard memory limits set on your VMs or ressource pools? If a VM needs more RAM than it's limit, then the memctl driver will try to inflate and reclaim physical RAM, forcing the guest OS swap on it's own pagefile.
Take a look at esxtop to check the TPS and balloning/swapping stats too. If you want to maximize TPS and thus memory usage efficiency, you might want disable large pages in ESX or (if you have a NUMA system) enable TPS across NUMA nodes (http://frankdenneman.nl/2010/02/sizing-vms-and-numa-nodes/).
Thanks MKGuy (+Others) .
Like I said, I'm guessing it's not a problem really, as when I enable DPM, vcenter shuts down one host. I guess that if I didn't have an anti-affinity DRS rule keeping 2 VMs seperate then it would shut 2 down, as my (recently-enabled) HA info says that I have capacity for 2 host failures (cluster is only 3 hosts...)
I was just a little concerned about VM Performance with ballooning memory.
-
VCP4 (but still learning
Please consider awarding points.