Linux Virtual Machine Process Scheduling

In the Linux VServer virtual machine (VM) facility, the allocation of processor time to multiple VMs is managed using a Token Bucket Filter (TBF). This system ensures that each VM receives a fair share of processing resources, particularly in environments with single or multicore processors.

Fraction of Processor Allocation

Processor Allocation Formula: The ratio of token generation rate (R) to the time interval (T) determines the fraction of a single processor's capacity allocated to a VM: 

R/T = Fraction of processor allocation

Linux VServer Token Bucket Scheme

For example, in a system with four cores, if a VM is intended to use the equivalent of one full core, R is set to 1, and T is set to 4.  

Overall System Limitation: The total fraction of processor time allocated to all VMs should not exceed the available processing power. If there are N VMs, the sum of their allocations must satisfy

Setting Parameters: S and M 

• S (bucket capacity) and M (minimum threshold) are parameters that regulate how a VM behaves after utilizing its burst time. 
• Burst Time (B) and Hold Time (H): After using the processor for a specified burst time B, the VM incurs a hold time H where it cannot resume operation until enough tokens accumulate. 
• The desired values of S and M can be calculated using the following formula:

M=W*H*R/T

S=W×B×(1−R/T) 

• W represents the scheduler's frequency (decisions per second).

Example Calculation:

• For a VM limited to 1/2 of the processor time, allowed to run for 30 seconds with a 5-second hold time, and with the scheduler running at 1,000 Hz: 

M=1,000×5×0.5=2,500 tokens 

S=1,000×30×(1−0.5)=15,000 tokens

Significance

These formulas and parameters allow for precise control over how much processor time a VM can use and how long it must wait before resuming after depleting its allocated tokens. This ensures fair resource distribution and prevents any VM from monopolizing the CPU. 

Conclusion 

The TBF mechanism in Linux VServer is critical for managing processor allocation in environments with multiple VMs. By carefully configuring the parameters R, T, S, and M, system administrators can control VM behavior during bursts of activity, ensuring that all VMs receive their fair share of CPU time without overwhelming the system.