SimGrid  3.20
Versatile Simulation of Distributed Systems

Detailed Description

Interface created to mimic IaaS clouds.

With it, you can create virtual machines to put your processes into, and interact directly with the VMs to manage groups of processes.


sg_vm_t sg_vm_create_core (sg_host_t pm, const char *name)
 Opaque type describing a Virtual Machine. More...
void simgrid::s4u::VirtualMachine::set_bound (double bound)
 Set a CPU bound for a given VM. More...

Function Documentation

◆ sg_vm_create_core()

sg_vm_t sg_vm_create_core ( sg_host_t  pm,
const char *  name 

Opaque type describing a Virtual Machine.

All this is highly experimental and the interface will probably change in the future. Please don't depend on this yet (although testing is welcomed if you feel so). Usual lack of guaranty of any kind applies here, and is even increased.

Opaque type describing a Virtual Machine.

The name of the VM must be unique among all hosts.

◆ set_bound()

void simgrid::s4u::VirtualMachine::set_bound ( double  bound)

Set a CPU bound for a given VM.

  1. Note that in some cases MSG_task_set_bound() may not intuitively work for VMs.

For example, On PM0, there are Task1 and VM0. On VM0, there is Task2. Now we bound 75% to Task1@PM0 and bound 25% to Task2@VM0. Then, Task1@PM0 gets 50%. Task2@VM0 gets 25%. This is NOT 75% for Task1@PM0 and 25% for Task2@VM0, respectively.

This is because a VM has the dummy CPU action in the PM layer. Putting a task on the VM does not affect the bound of the dummy CPU action. The bound of the dummy CPU action is unlimited.

There are some solutions for this problem. One option is to update the bound of the dummy CPU action automatically. It should be the sum of all tasks on the VM. But, this solution might be costly, because we have to scan all tasks on the VM in share_resource() or we have to trap both the start and end of task execution.

The current solution is to use setBound(), which allows us to directly set the bound of the dummy CPU action.

  1. Note that bound == 0 means no bound (i.e., unlimited). But, if a host has multiple CPU cores, the CPU share of a computation task (or a VM) never exceeds the capacity of a CPU core.