I would like to determine the peak CPU usage for a VM in the custom UI. How would I do this?
If I'm using a supermetric, there is no attribute kind under CPU entitled "peak usage".
You certainly get points for asking a long question! Although to be fair, it's probably my fault for feeding you so much info to digest at once.
1. I would say you can do it two ways visually. Use the Operations>Detail view to get the "normal ranges" and use the upper limit on the normal range as your typical max. Or of you want an absolute peak, add the metric to a graph and observe the highlighted high and low watermarks.
2. There are lots of ways to view data, but if you want to view a single metric over a time period, we could use the following widgets [off the top of my head]:
Top-N Analysis
Weather Map
Data distribution analysis
Of these, I would think of your desire to get a "peak". From that you want an actual value, so that eliminates the Weather map. Then I think about the concept of a "peak" and what is the next best thing, I think 95th percentile.. then I come up with the data distribution widget. Ultimately the decision is yours given your particular use case.
3. You can have it apply to multiple VMs - it is just a widget capable of being a self-provider or a receiving widget (select metrics, NOT selected resources). Each resource/metric will add to the widget and you can remove them if you like after being added. You can certainly just the time period you're viewing.. it's the standard date/time options you have in all the other graph widgets. With the data distribution widget you want as much time as possible to get an accurate histogram of the workload cycles.
4. This is similar to the question above. You can certainly add individual resource metrics or a SM that sums up or avgs groups of resources. And as above, the time period is configurable just like the other graph widgets.
5. This is part of Q4 - you can sumN or avgN up any number of resources with a SM, then view that SM in the data distro widget. When I said container, I mean an application, tier, group, resource pool... anything that acts as a parent container within vCOps.
6. That is correct, SMs calc using the last-collected data points.. aka single time period. When you refer to the longest time-period, I was giving you some examples of the attributes that are comprised of multiple time periods that are calculated and created by behvaior of the VMware vCenter adapter. In this case, you've got the 1,5,15min CPU avgs. These specific simple moving avgs cannot be specific OR duplicated using SMs. The closest you'll get is finding a way to distribute what you need via a widget [at this time].
7. The data distro isn't for exporting, it's for visual functions. If you want the data in CSV, just use a typical metric graph and export to a csv and calc your own peaks, avg, 95th, whatever.
8. This is a bit of a mix of all of your questions put together.
You are correct if you added those 3 VMs, then calc'd the 95th of the avg. Like Q7, you wouldn't export the data from the data distro because the raw data points aren't displayed - you'd want to move the metric to a standard metric graph and export via csv.
Enjoy..
A 'peak' suggests you have multiple values to choose from. I would usually recommend using the visual drilldown to calc to normal ranges, but if you want a different way to do it, use a data distribution analysis widget. Feed it your CPU Demand (Mhz) or (%). Calc the widget to show the 95th percentile, then you'll see your utilization with the top 5% of garbage removed... I would use a 95th percentile long before a single "peak" data point. It's not for everyone, but it's nice for some things (IOPS!).
You can get 1min,5min,15min peaks for CPU, but those metrics might not be of the type you're looking for (throttled/running/etc).
Hi
Welcome to the communities.
There must be CPU Metrik
OK thanks for the input. I think the goal here is to look backward over a history of the last 3 months, and see what peak usage was for CPU for a VM. What happened at the end of the month when the company did its financials and demand spiked so we can prepare for it next month? The idea was to use the Custom UI to group a set of VMs together from different datacenters and clusters and see the average among those VMs. What is the most effective way to set this up?
You can average across multiple resources during a single time period (last collected), and is accomplished using the avgN() formula. And that would give you the avg for all of the resources in that container to which the VM is applied.
As we've reviewed already for some of your other questions, you can't average for a single resource over multiple time periods using a supermetric - the capability isn't there.
If you wanted to do a dashboard for this, you can use some variety of graphs and eye-ball an average, or go with a data distro widget.
If you stop and think about what you're trying to do, "avg" might not be best for you. If I had a group of resources for an application or some other type of container, and assuming the total cpu demand isn't already available, I'd want to have a SM that sumN the CPU Demand (Mhz). Then if I want to see the total demand for the group, just open that SM and graph it or pipe it in to a widget.
Thanks again for all the input. I will try to respond to both of your posts inline. Thanks for your patience with my questions!
A 'peak' suggests you have multiple values to choose from. I would usually recommend using the visual drilldown to calc to normal ranges, but if you want a different way to do it, use a data distribution analysis widget. Feed it your CPU Demand (Mhz) or (%). Calc the widget to show the 95th percentile, then you'll see your utilization with the top 5% of garbage removed... I would use a 95th percentile long before a single "peak" data point. It's not for everyone, but it's nice for some things (IOPS!).
Good suggestion for how to define "peak". So I’ll start by switching from attempting to get a "single peak point" to attempting to get the 95th percentile as you suggested. With that established, I now I need to try to clarify a few things on your recommendations:
1. You mentioned that
“I would usually recommend using the visual drilldown to calc to normal ranges,”
When you say “visual drilldown” what specifically are you referring to? Using a widget in the Custom UI, or using the vSphere UI and drilling down and In what screen specifically would I be doing this?
2. You mentioned that:
“but if you want a different way to do it, use a data distribution analysis widget.”
I’m not looking to be different necessarily but for the most effective way to do it, whatever that is, with this goal:
- I need to be able to view get to where I can view the 95th percentile as you said for a group of VMs across different vCenters, datacenters, clusters and sites. I need the flexibility to be able to group a specific set of vms together that may not natively be all together in the same cluster or folder.
Is data distribution analysis widget the best method for that? Is there a better method?
3. Concerning this step:
Feed it your CPU Demand (Mhz)”
If I did use the data distribution analysis widget and fed it my CPU Demand, then:
-Can I make this apply to a hand-picked group of VMs as opposed to just a single VM?
-Can I make the time period over which I am viewing to be for the last 3 months?
4. You mentioned that:
“You can average across multiple resources during a single time period (last collected)”
In this case, say the resources I am concerned about are all VMs running Windows 2008 STD R2 in any cluster or datacenter, and there are 3 of them spread abroad. If I am doing the average over a single time period, can that time period be the last 3 months or is there a limitation to what the "time period" length is?
5. You mentioned that:
and is accomplished using the avgN() formula. And that would give you the avg for all of the resources in that container to which the VM is applied.
When you say “container to which the VM is applied” - where do I apply a VM to a container?
6. You mentioned that :
As we've reviewed already for some of your other questions, you can't average for a single resource over multiple time periods using a supermetric - the capability isn't there.
OK so a supermetric can be used only over a single time period meaning it is only for a short term, present tense, “Just happened or happening now” visibility and not for something like the last 3 months? If I understand correctly the longest time period was something like 15 minutes and can’t be made longer? If so that would rule supermetrics out of anything that requires much of any history behind it.
7.You mentioned that:
If you wanted to do a dashboard for this, you can use some variety of graphs and eye-ball an average, or go with a data distro widget.
In this case I need to be able to do things like export the results such as last 3 months of data to csv. So is data distro widget my best option?
8. You mentioned that:
If you stop and think about what you're trying to do, "avg" might not be best for you. If I had a group of resources for an application or some other type of container, and assuming the total cpu demand isn't already available, I'd want to have a SM that sumN the CPU Demand (Mhz). Then if I want to see the total demand for the group, just open that SM and graph it or pipe it in to a widget.
So here is a better example of what I am trying to do. There are 3 virtual machines I am concerned about.
VM1 had a 95th percentile rating of 1.2GHz for the last 3 months
Vm2 had a 95th percentile rating of 1.4 GHz for the last 3 months
Vm3 had a 95th percentile value of 1.6 GHz for the last 3 months
I group these VMs together, and I see that the average of the 95th percentile rating for them is 1.4GHz for the last 3 months. I export this to csv. Of the things we have discussed, what is the best method to do this?
Thanks!
You certainly get points for asking a long question! Although to be fair, it's probably my fault for feeding you so much info to digest at once.
1. I would say you can do it two ways visually. Use the Operations>Detail view to get the "normal ranges" and use the upper limit on the normal range as your typical max. Or of you want an absolute peak, add the metric to a graph and observe the highlighted high and low watermarks.
2. There are lots of ways to view data, but if you want to view a single metric over a time period, we could use the following widgets [off the top of my head]:
Top-N Analysis
Weather Map
Data distribution analysis
Of these, I would think of your desire to get a "peak". From that you want an actual value, so that eliminates the Weather map. Then I think about the concept of a "peak" and what is the next best thing, I think 95th percentile.. then I come up with the data distribution widget. Ultimately the decision is yours given your particular use case.
3. You can have it apply to multiple VMs - it is just a widget capable of being a self-provider or a receiving widget (select metrics, NOT selected resources). Each resource/metric will add to the widget and you can remove them if you like after being added. You can certainly just the time period you're viewing.. it's the standard date/time options you have in all the other graph widgets. With the data distribution widget you want as much time as possible to get an accurate histogram of the workload cycles.
4. This is similar to the question above. You can certainly add individual resource metrics or a SM that sums up or avgs groups of resources. And as above, the time period is configurable just like the other graph widgets.
5. This is part of Q4 - you can sumN or avgN up any number of resources with a SM, then view that SM in the data distro widget. When I said container, I mean an application, tier, group, resource pool... anything that acts as a parent container within vCOps.
6. That is correct, SMs calc using the last-collected data points.. aka single time period. When you refer to the longest time-period, I was giving you some examples of the attributes that are comprised of multiple time periods that are calculated and created by behvaior of the VMware vCenter adapter. In this case, you've got the 1,5,15min CPU avgs. These specific simple moving avgs cannot be specific OR duplicated using SMs. The closest you'll get is finding a way to distribute what you need via a widget [at this time].
7. The data distro isn't for exporting, it's for visual functions. If you want the data in CSV, just use a typical metric graph and export to a csv and calc your own peaks, avg, 95th, whatever.
8. This is a bit of a mix of all of your questions put together.
You are correct if you added those 3 VMs, then calc'd the 95th of the avg. Like Q7, you wouldn't export the data from the data distro because the raw data points aren't displayed - you'd want to move the metric to a standard metric graph and export via csv.
Enjoy..
Ok great thanks for the input.
