Hi ESABO12, I'm glad to hear that the whitepaper was helpful, and also very interested in the follow-up questions about how to get sizing right. These are topics that we're looking to cover in the upcoming whitepapers that will be out soon (I'll post links to this thread when they're released). We're also planning a free webinar on the topic with vExpert Greg Shields on June 2 (sign up link here), who is very knowledgeable in this area and will be taking questions throughout the webcast.
What we've found is that even with knowing which metrics to use to determine sizing, the process can still involve some non-technical factors moreso than just baselining and establishing how much capacity is left that can change allocation amounts such as SLA provisos, expected application usage growth, and allaying concerns from application owners, etc... which can make sizing a bit more of an art. Here are some rules of thumb that can help in baselining and in figuring out how to size correctly (these will be explained in more depth and technical detail in those whitepapers that we are currently working on):
Determine What an Appropriate Time Frame is for Evaluation - Some applications may only be used once a month or once a quarter, others get the same use every day. To get an accurate assessment of how many resources are necessary, the application's consumption should be measured from end to end in what would be a usage cycle for the application.
Base Sizings on Peak Usage Periods - When the right time frame is determined, look for the peak amount of usage, as any VM's resource allocation needs to have enough "juice" to maintain high performance when an application is at maximum usage.
Add a Buffer - After finding the peak, add on a percent more than what is observed from past performance data so that if a future peak is higher than an observed peak, the application will maintain its performance... That percent to use as a buffer is probably not a "one size" fits all, and figuring out for each application what it should be is where things become a bit more of an art.
Start with the Minimum Amount of Resources That Will Keep Good Performance for a VM out of the Box - This approach can help maintain appropriate sizings without causing undue performance concerns: It's a lot easier to tell an application owner that you're giving their apps more resources, than finding out that they were vastly over-allocated, and then telling them that you need to take something back.
Ping the Application Owners for Growth Data - If the application owners can give a good idea of which applications will grow in user counts, will experience updates that could make the app more or less resource-hungry, or will start to experience more usage, that VM should be closely watched to make sure it will have enough resources to maintain high performance.
Run a Sizing Assessment Process As Often as Possible - Because application usage can change frequently, sizing needs for the resources that maintain that app's performance may also need to change. That means that the baseline will always be changing - and it needs to be reassessed as often as possible to ensure that the VM continues to have what it needs, and is not needlessly taking up resources that other VMs could use.
As for CapacityVIEW, it just runs on a Windows workstation and does not integrate into the vCenter client. We designed it as a quick to use assessment tool which was just a few MB download. Our commercial product which works on VM sizing, Optimizer does launch from within the vCenter client, and will analyze the metrics described in the whitepaper to assess VM allocations over the entire environment in a regularly recurring automated fashion - finding the peak usage and suggesting sizing recommendations that can be scheduled or implemented automatically through the vCenter API. Many of our clients also use the graphical sizing representation per VM to explain to application owners that even at peak usage, their applications are using a small fraction of what they've been allotted (I've attached a screenshot of what that looks like and highlighted where that is shown). Optimizer installs as a virtual appliance in about 20 minutes, so you'd be able to get that automated baseline set up quickly if you'd want to try it out.
I hope that this gives some more insight, and in a few weeks, I'll post the whitepapers that we plan on releasing which will have more detail. Let me know if there are any other areas
Alex Rosemblat
VKernel
