Mitigating Cold Start Problem in Serverless Computing with Function Fusion

Area of Science:

• Computer Science
• Cloud Computing
• Artificial Intelligence

Background:

• Serverless computing is crucial for cloud-based AI services due to its simplicity and scalability.
• The cold start problem in serverless computing introduces latency, impacting workflow performance.
• Function fusion is a potential solution to reduce cold start latency.

Purpose of the Study:

• To mitigate cold start latency in serverless workflows using function fusion.
• To develop a workflow response time model that accounts for latency factors.
• To optimize function fusion strategies for parallel execution scenarios.

Main Methods:

• Identifying key latencies affecting workflow response time.
• Developing a workflow response time model incorporating identified latencies.
• Proposing an efficient method to find optimal function fusion solutions.

Main Results:

• The proposed function fusion approach significantly reduces workflow response time.
• Response times were reduced to 28-86% of the original workflow response times across five test cases.
• The method effectively balances cold start mitigation with considerations for parallel function execution.

Conclusions:

• Function fusion is an effective strategy to address cold start latency in serverless AI workflows.
• The developed response time model and optimization method provide a framework for improving serverless application performance.
• This research contributes to more efficient and responsive cloud-based AI service delivery.