Episode 78: Fault Tolerance with Bob Hanmer Pt. 2

Load Shedding

Load shedding is a crucial strategy in managing system overloads by ensuring that systems operate within their processing capabilities. Bob Hanmer explains that slowing down processes can help maintain system sanity by limiting the workload to what the system can handle, thus avoiding unrecoverable overload situations 1. This approach is particularly beneficial in scenarios where human users might repeatedly submit requests due to impatience, as seen in telephone systems 1. Priyanka Raghavan appreciates the value of these patterns, noting how they expand the toolbox for software architects and developers 2.

Just slow it down. Just because by doing that, you keep the system sane, you are limiting what you're doing.

--- Bob Hanmer

These patterns, such as fresh work before stale, are derived from extensive telecom experience and are widely applicable in various overload handling techniques.

   

Resource Queuing

Resource queuing serves as a mitigation strategy during overload conditions by managing incoming requests efficiently. Bob Hanmer describes how queuing requests can smooth out spikes in demand, allowing processors to handle tasks more effectively 3. This method requires equitable distribution of resources to ensure all users receive good service, as defined by system requirements 3. In cases where shedding load isn't feasible, sharing the load with other system components can be an alternative solution.

Sometimes you can get away with queuing requests as they come in to process them a little later, because it might be a very momentary spike.

--- Bob Hanmer

By planning these strategies at the design stage, systems can better allocate resources during periods of overload, whether locally or across networks.