Ability to exlude VMs from Analysis and Reporting
Would be great to have the ability to exclude VMs from the analysis and report outputs as we can with physical hosts. I have customers who have asked for VMs to be discounted from sizing and there is no current way apart from asking the customer to scan individual guest OS's. If this feature could be added would be great if it could accept a csv/xls type input as well and individual exclusions in the web interface as the numbers of VMs to be excluded can often be quite high.
This feature is actually under construction. It will be called Virtual View. It will be a new tab in the Optical Prime Service. An exact date has not yet been identified, but we anticipate it will be out sometime in April.0
Is this option already available to use?0
Still not available?1
Hi Dave, I tried today and that is still not available. So sad.0
Today I have also had a request from a customer for this feature. Please could you add an update on when this will be available. Thanks0
Apologies upfront, I didn't mean for this to get this long. Hopefully, this helps with an understanding of how interconnected this all is and how much it helps to understand the outcome desired as the most important piece of data to a request.
This is a deep feature request. You would really need to understand the depth of the ask to appreciate the comment. You should ask your customer what outcome they expect if they had that feature.
You can exclude nodes (and all their VMs), clusters (and all their VMs), servers, disks, etc... these things effect the entire project and they can successfully be removed to alter the aggregate summaries.
For example, if you had a 3 node cluster and you dropped one, 1/3rd of the resources are now gone. That is a calculable outcome as those records can be purged.
So, for a feature like excluding a VM... you need to something to exclude it from and you need the data to exclude!
You can't exclude it from a node and assume it's impact on the node summary (like IOPS over time). Those correlating records don't exist. That would be like tell me your car gets 20mpg and you want me to remove the mirrors and then asking me what your MPG improvement would be. I would need a lot of data to give you the answer, but I can't knock of your mirrors really easy :)
You could do memory and capacity summaries based on assignments for static values, but you could not update the node performance graphs accurately.
So, if you had a node that ran 100 VMs and had a used capacity of 50GBs and you removed 75 of those VMs. Yes, we could easily calculate the loss of capacity. That data exists.
But..., if we had a node that ran 100 VMs and had a 95th % of 4,300 IOPS and you removed 50 of the VMs. What do we do with the performance graphs? We don't have the data points to update the performance graphs and we don't have them because VMware doesn't have them.
So, you would have a fragmented recalculation and the action of removing a VM and its outcome would be greatly misunderstood. It's reasonable that a person excluding half their VMs would expect to see, maybe, half the performance in the graphs. But it can't happen. Not today anyway.
So, we have been busy building really valuable things that are related to VMs where removing a VM would actually be understood. This is the underpinning for a feature like this.
Just last month (Dec of 2020) Virtual Tab now aligns to Performance tab. So, if you drop a node, then all the VMs it's running also drop from Virtual Tab. That is a HUGE step forward in real time data manipulation.
Therefore, the VM distribution graphs, the provisioning contrast graphs, cloud pricing, excel, etc... ALL now follow the save state of Performance tab. That is massive work!
These richer VM level features can benefit from alterations at the VM level since they deal with VM level data.
I often get something like "I want to drop a VM for my encompass report", which is fine. But one has to remember that LO deals with much more than VMware data and has a depth of presentation that far exceeds a simply summary of static values. Changes we make have to be compatible across the board.
The Live Optics presentation layer is largely built around displaying data related to performance over time vs summaries of configuration data.
So, as a product team we have to make sure that one addition doesn't create confusion in other areas. Once you start peeling back the onion, it's crazy deep.
What would be awesome is if you documented your customer's ideas on what they want achieved or even as an SE what you want achieved with a feature like this.
The short version of this is that there was almost a year of underpinning work to get to where a feature like this makes sense. I would expect that this wouldn't creep into the project until mid-FY22 as the outcome desired becomes more understood (and knowing the cloud roadmap ahead of us, which is the highest priority).0
Much appreciate the continued efforts on this after I requested it 2 years ago. I did not understand the challenge it would create.
'You should ask your customer what outcome they expect if they had that feature.'
In response to this the customers want to know what resources are necessary to run a subset of their workloads. As SEs we are often working with customer who are changing their workloads as they are changing their platforms, they want to decommission VMs when they migrate, this is more often the case than not. So, having the ability to remove VMs (its normally the SE not the customer doing this in LO) lets us show the resources for the target virtual workload that we can then size for.
Although it would be helpful to exclude based on nodes or other hardware based filters, it would be more helpful to exclude solely on VM as often the VMs to be excluded are likely to be scattered throughout the hardware estate, from my view I would much prefer a filter per VM than any other but any capability would be appreciated.
I don't understand the depth of detail you have mentioned, but I had thought that we could get individual IO profiles from VMs, so I am confused by not being able to take that away from the total, although I do see these are statistical summing calculations rather than all integer operations. Even if we can get everything some general high level details would be useful, total disk capacity, used VM memory for target VMs etc.
Again, thank you for your engagement and efforts on this they are very much appreciated and I am sure the SE community would benefit greatly from having this capability.
Hi Sam and Philip,
I think all of this discussion is benefit, to ever evolve our tools.
And I agree about what Philip said. Many times we, as SE, need to ignore some VMs, due to descomissioning or even due to unnecessary in a DR project, for example.
About individual IO profiles, I think it's possible to implement this feature since in the export file we have the tab "VM-Performance", that is exactly the individual IO profiles about each VM. So, if we have the date individualy and the total, I think it's okay to calculate the one less the other.
And, as Philip said, thank you for your engagement and efforts on this they are very much appreciated and I am sure the SE community would benefit greatly from having this capability.
In regards to that.
We have a VM performance value...
- but only if the person elects to gather VM performance (it's optional)
- only if they are running VMware
- only if they running VMTools in the guest
- only if the VM is running itself. (VMtools doesn't report if the VM is off, but this is OK since it didn't contribute to the hypervisor performance anyway)
So, if any of these qualifications aren't met, we have to still make it work. LO supports KVM, Hyper-V, etc.. and all in the same or even a mixed interface.
Even with VM performance, we get single data points (peak and average) and no time assignment to when that occurred. To deduct performance from things like 95%, the graphs, etc.. correctly, you would need telemetry data that was equal to the time frame of the collected data on the hypervisor. So, these data points will still not help with performance graphs or overall IO summations (you can not add the VM performance data points from excel and assume the IO need).
Regardless, we hear you loud and clear and we are adding it. It is just more than it seems on the surface, but we will add controls to allow the opting of VMs and allow that to have an effect on Virtual Tab, excel and Cloud Pricing only. We will leave the Environment View and Performance Tabs (which look at the Hypervisor data) alone.0
I asked the same question to LO support. They said to do this: Optical Prime Bulk Server Import with Live Optics .NET Collector – Live Optics Support
And specify every VM IP address. I haven't tried it but the customer is requesting a new platform for 400 VMs out of 1000; partial migration.1
Hi Bill Lipman,
Appreciate the info on the Bulk Server import - did you ever run that successfully? And if so, what did the format of the server input file look like, was it one IP per line, or did you use a delimiter (and if so, which one)?
Appreciate your feedback, I have a customer with two monstrous TSM backup servers that are skewing the entire analysis that I want to exclude. Your post is giving me hope :-)
The customer's VMs were VDI persistent desktops so we're working on estimating what was needed. With over 400+ VMs the account team thought it would be too much work for the customer to do. Would still be nice to filter by VM once the LO capture is uploaded. Really hope that becomes a feature someday in LO.
Plan B could be to move all the VMs to specific hypervisors and filter by host in LO as an option, if the customer is OK with moving VMs around.0
Thank you for the feedback, Bill! I'm also hoping for an automated way of excluding VMs, but am also exploring if I can have the templates to the AIR output so I can work through things manually in the meantime. I realize excluding VMs has a lot of implications on the performance data side (that Sam has outlined above), but in my case my customer is more interested in understanding their capacity needs vs. analyzing performance.
Appreciate your help!0
Bill Lipman - appreciate your feedback! I'm going to have to strip out the numbers manually, thankfully I only have two servers that are skewing the AIR reports. Can't wait for this to be available as part of Live Optics!0
Hello guys, anyone have an alternative to do this filter per VM?
I collected LO, and now need to ignore some VMs that will be migrate to EC2, but anothers not.
Still waiting for this awesome feature.0
Best bet at the moment is to extract RAW data to spreadsheet and manually filter.0
Please sign in to leave a comment.