How to do Capacity Planning based on LAB Performance/Load Tests?

[Vaibhav]

Hi,

I am looking for answer or procedure we should follow to do the capacity planning?

# Let me put the question in a different way,
suppose customer is asking for number of servers and configuration i should use based on different user base.How do i calculate? or Answer this?

#This is a common question but not getting proper answer for this, request you to please share any reading material or link which i can use to study actual formula of Capacity Planning.

Thanks

 

[AnmolD]

@Vaibhav Ther is no straight forward answer to this.
I have seen people doing sizing (in one of my previous company) mostly by experience. Ideal method is to design an accurate workload and then based on the results of testing, predict the size of your hardware.
Though workload design for a new application would be an approximation. A typical methodology is shown below

 

[davecb]

If you work in time units you can do a mathematical model (a queuing network) of how long it takes to do a unit of work, and that can be scaled for different hardware. Working with %cpu or memory sizes and scaling linearly will give you a model that will (naturally!)predict straight lines, while the actual time-vs-load curve is a hyperbola, and looks like "_/" when plotted. See anything by Neil Gunther or Teamquest for the real skinney.

 

[admin]

@devecb can you please give me a pointer to Neil Gunther? Do you have a ebook that we can refer

 

[Krishna18in]

Extrapolation doesnt always work, There is no silver bullet formulae to calculate the Extrapolation of results.
How can we expect that 1 server can handle 100 users load ==> 10 servers can handle 1000 users load.

 

[anmoldubey]

How can we expect that 1 server can handle 100 users load ==> 10 servers can handle 1000 users load.

That defines horizontal scalability and is never directly proportional !!