There is not a direct answer for that. The computacional complexity of each web request depends on what the real server application is doing. Furthermore, the time to respond those requests depends on multiple factors, including: the computacional capacity of the physical machine/virtual machine, the number of users currently accessing that server, etc. If the server application is just providing static web pages, the process is simple and fast. If it is querying a database to build a dynamic page or performing other computation, the process is more complex.If each Cloudlet represents the process that responds to a specific request, there may be multiple ways to define the Cloudlet length. It depends on which of those characteristics I mentioned you want to simulate.To model such a kind of application, you may rely on some datacenter trace file. Those traces can provide the time each request took to be answered by the server. If there is such an information on the trace file, you can compute the cloudlet length easily. For instance, consider that:

• one particular request took 0,5 second to be answered;
• your Cloudlet is using 1 CPU core;
• it is running into a VM having at least 1 core available for that Cloudlet;
• and each VM CPU core has 1000 MIPS capacity.

That means each VM core can process 1000 Million Instructions per second. This way, to make that Cloudlet to run for 0,5 second, we can define its length as 500 MI. For that scenario, the length of your Cloudlets can be computed as VM_CPU_MIPS * TIME_TO_PROCESS_THE_WEB_REQUEST.I hope it helps. If you want to use a more accurate, full-featured, fully documented and highly extensible framework, try CloudSim Plus at http://cloudsimplus.orgI please don't forget to give the project a star to help promoting it.