Tell me how much to provision to reach my desired state.

Idea created by ryan.hinkamper on Nov 5, 2015
    For future consideration
    • Rene Van Den Bedem
    • nikop
    • fgerber
    • gilesogram
    • byemini
    • battybishop

    I have a cluster telling me "Critical Mem congestion, provision more physical machine capacity" to reach my desired state. Both of the hosts below are in the same cluster.


    It's good that I am being notified what the problem is, but why not tell me how much I need to resolve it? I created a template mocking up these hosts with more memory and ran a plan to see what my cluster would look like with the memory upgrade and found what I needed to reach the desired state in question.


    Provision more capacity.JPG



    I ran a plan for what it would look like if I removed the two existing hosts, and added two new hosts that were identical except I increased their installed memory based on a supported configuration per the vendor spec sheet. In the picture below you can see the utilization index drops dramatically.



    In the following picture here you can see where it shows the two old hosts removed and two new ones added.


    This shot shows in detail that current memory utilization is very high.




    This last shot shows the drastic impact of the hosts with increase memory capacity under normal workload.



    While going through all these steps to find the resolution to the issue is not hard, why not just tell me how much memory I need to add to my "marketplace" to bring prices down and create my good state? To take it a step further it would be nice if I could input a vendor sku of my host's in production, and the system could tell me exactly what a supported memory configuration is for my environment, but that may be a bit of a pipe dream. At the very least it can tell me how much more I need in raw numbers, then I can take that and figure out what my hardware will support. Then I can decide if I need more host's to support what's needed or simply add to what I have.

    One could argue that the template I created had more memory than I needed based on the final utilization numbers, and they may have a point. To better fine tune those numbers I would have to run this report over and over with different memory configurations to find the "sweet spot", but it all goes back to my initial proposition, why not just give me that number to begin with? The same scenario would be in play with any of the commodities in the market (CPU, IO, NET).