Network Load Balancing Methods: Experimental Comparisons and Improvement

Shafinaz Islam

DOI: https://doi.org/10.48550/arXiv.1710.06957

2017-10-19

Abstract:Load balancing algorithms play critical roles in systems where the workload has to be distributed across multiple resources, such as cores in multiprocessor system, computers in distributed computing, and network links. In this paper, we study and evaluate four load balancing methods: random, round robin, shortest queue, and shortest queue with stale load information. We build a simulation model and compare mean delay of the systems for the load balancing methods. We also provide a method to improve shortest queue with stale load information load balancing. A performance analysis for the improvement is also presented in this paper.

Distributed, Parallel, and Cluster Computing,Performance

What problem does this paper attempt to address?

This paper attempts to solve the problem of how to optimize the load - balancing algorithm when using obsolete state information in network load - balancing. Specifically, the paper studies four load - balancing methods: Random, Round Robin, Shortest Queue, and Shortest Queue with Stale Load Information. The author constructs a simulation model to compare the average delay of these methods and proposes an improved Shortest Queue method, namely Shortest Queue with History, to improve performance when using obsolete state information.

### Main research questions:
1. **Performance comparison of different load - balancing methods**:
- The paper experimentally compares the performance of the four methods of Random, Round Robin, Shortest Queue, and Shortest Queue with Stale Load Information, mainly focusing on average response time and resource utilization.

2. **The impact of obsolete state information on load - balancing**:
- The author explores the performance degradation problem of the Shortest Queue method when using obsolete state information and analyzes the impact of different stale periods on system performance.

3. **Improving the Shortest Queue method with obsolete load information**:
- A new Shortest Queue method with History (HSQ) is proposed and its effectiveness is verified in experiments. This method improves the average response time of the system by recording the task history scheduled during the obsolete period.

### Experimental design:
- **Simulation tool**: Use CSIM20 for simulation modeling.
- **Performance indicators**: Mainly focus on average response time.
- **Experimental factors**:
- Load - balancing strategies: Random, Round Robin, Shortest Queue (USQ), Shortest Queue with Stale Load Information (SSQ), Shortest Queue with History (HSQ).
- System load: From 0% to 90%.
- Length of obsolete period: From 1 second to 25 seconds.

### Experimental results:
- **USQ vs. Random and Round Robin**: USQ shows the best performance under all loads, while the performance of Random and Round Robin strategies is poor.
- **SSQ vs. Random**: As the obsolete period increases, the performance of SSQ decreases significantly and is even worse than the Random strategy in some cases.
- **HSQ vs. SSQ**: HSQ is superior to SSQ under all loads and obsolete periods, especially under high loads and long obsolete periods, and the performance improvement is more obvious.
- **HSQ vs. USQ**: Although the performance of HSQ is not as good as that of USQ, the gap is relatively small, and under extreme obsolete periods, the performance of HSQ is close to that of the Round Robin strategy.

### Conclusions:
- **Optimal strategy**: USQ performs best when using the latest state information.
- **Improved strategy**: HSQ significantly improves performance when using obsolete state information and is suitable for load - balancing problems in actual systems.

Through these studies, the paper provides valuable references and improvement directions for load - balancing in actual systems.