How do you design a fair and efficient scheduling policy for multiprocessor systems to avoid starvation?

1 What is deadlock?

Deadlock is a situation where a set of processes are waiting for each other to release some resources, but none of them can proceed. For example, suppose there are two processes, A and B, and two resources, R1 and R2. A holds R1 and requests R2, while B holds R2 and requests R1. Neither of them can continue until the other one releases the resource they need. This is a deadlock.

2 What is starvation?

Starvation is a situation where a process is indefinitely delayed from accessing a resource or executing its code, because other processes with higher priority or more frequent requests are always favored. For example, suppose there are three processes, C, D, and E, and one resource, R3. C has the highest priority and always requests R3, while D and E have lower priority and request R3 occasionally. C always gets R3 and never releases it, while D and E never get a chance to use it. This is starvation.

3 Why do they happen?

Deadlock and starvation can happen due to various factors, such as the number and type of resources, the allocation and request policies, the communication and synchronization mechanisms, and the workload and behavior of the processes. For example, deadlock can happen if there are circular dependencies among the resources, if the processes request resources in a fixed order, or if there is no preemption or timeout mechanism. Starvation can happen if there is no fairness or balance in the resource allocation, if the processes have different priorities or frequencies, or if there is no feedback or adaptation mechanism.

4 How to prevent deadlock?

One way to prevent deadlock is to avoid the four necessary conditions for deadlock, which are: mutual exclusion, hold and wait, no preemption, and circular wait. For example, you can use lock-free or wait-free algorithms that do not require exclusive access to resources, or you can use timeouts or interrupts that can release resources if a process is blocked for too long. Another way to prevent deadlock is to use deadlock detection and recovery techniques, such as using a graph or a matrix to represent the resource allocation and request status, and applying algorithms to find and break cycles or deadlocks.

5 How to prevent starvation?

One way to prevent starvation is to use a fair and efficient scheduling policy that can allocate resources to processes according to their needs, priorities, and frequencies. For example, you can use a proportional-share or a lottery-based policy that can assign resources based on the proportion or the probability of the processes' requests, or you can use a feedback-based or an adaptive policy that can adjust the allocation based on the performance or the behavior of the processes. Another way to prevent starvation is to use load balancing and resource sharing techniques, such as using a global or a distributed queue to distribute the requests among the processors, or using a shared or a replicated resource to increase the availability and the throughput of the resource.

6 How to design a fair and efficient scheduling policy?

To design a fair and efficient scheduling policy for multiprocessor systems, you need to consider several factors, such as the resource characteristics, the process characteristics, the system objectives, and the system constraints. For example, you need to know the type, the number, and the availability of the resources, the priority, the frequency, and the demand of the processes, the goals of the system, such as maximizing the utilization, the throughput, or the response time, and the limitations of the system, such as the overhead, the complexity, or the scalability. Based on these factors, you can choose or design a suitable scheduling policy that can avoid deadlock and starvation, and achieve the desired performance and correctness of the system.