Hi,
One of the things I am finding with vCOPS is that a lot of the stuff I thought I knew is now going out of the window.
Before vCOPS we relied on vCenter and when the host started to report 80 or 85 % memory usage it was time to add more RAM.
It seemed like many people used this figure as a baseline.
vCenter is now reporting that my hosts are sitting at 90 and I get the occassional alarm.
Normally I would have placed an order for more hardware. I have this VMware vCOPS presentation that is telling me that whatever vCenter says is in accurate.
vCOPS is telling me (under details > workload) that my memory workload is like 32%.
The Memory|Usage counter shows the host memory like vCenter does but memory| demand is actually 32% of that and that is the number I need.
On the one hand it is cool that all is green in vCops, on the other hand I am nervous because vCenter is alerting.
Do I need to start ignoring vCenter now that I have vCOPS running?
Cheers
Well, I said the issue of memory management is complicated. We try to make vC Ops easy to use, but obviously some of the complications come through.
At any rate, yes, I think it's ok to ignore what VC says. Given that you're always run things a certain way according to VC, it's probably best to start overcommitting slowly. (Evolutionary change is always better than revolutionary. ) But the general answer is that yes, the VC alerts are in general much too conservative and aren't really taking a variety of performance data into account.
Good questions and observations.
The bottom line is that vCenter useing consumed values and vC Ops we use Demand based vaules. Lets look at memory Demand. It is based on Active memory and not consumed. Thus that is why you see the differance.
Based on the demand view in vCOPS everything is fine but then I notice that the host is actually ballooning. This tells me there are memory constraints and very confusing 🙂
AllBlack, it's a long and complicated issue. I've tried to summarize it in a VMworld talk I did on memory management: http://www.youtube.com/watch?v=YKaUtoQrLjo.
But the bottom line is that some memory reclamation is not bad for performance. A little bit of ballooning will generally have no performance impact. I think a lot of people avoid memory reclamation like the plague, but the reality is that a little bit is not all that bad, especially if you're implemented all the best practices around memory management (my talk covers some of them).
As Hugo mentioned, VC takes a "consumed" view of memory, which means that your consolidation ratios will be very low and you're not making the most efficient use of your memory. vC Ops uses a "demand" model, showing what the VMs are actually using rather than the amount of RAM that ESX has allocated each. Because of this, it means there will be some memory reclamation to shift the allocated RAM between VMs. But this is a good thing - memory is being moved to the VMs that need and deserve it most. And this can be done without catastrophic performance impacts (as many people worry about!).
Hope that helps.
Hi Kit,
I started watching your video and hope to finish it soon.
We always avoided oversubscription and also give the VM way too much RAM.
vCOPS is a great tool but as you said somewhat complicated.
In short, Can I ignore what vCenter is saying ? Hosts were sitting at 94% the other week. Normally we'd be running to the shop for more RAM now 🙂
Well, I said the issue of memory management is complicated. We try to make vC Ops easy to use, but obviously some of the complications come through.
At any rate, yes, I think it's ok to ignore what VC says. Given that you're always run things a certain way according to VC, it's probably best to start overcommitting slowly. (Evolutionary change is always better than revolutionary. ) But the general answer is that yes, the VC alerts are in general much too conservative and aren't really taking a variety of performance data into account.
Thanks Kit.
I'd like to say that your talk on memory management @ VMworld 2011 and the VC Memory Statistic Definitions document are brilliant.
I think I finally start to understand it 😉