## Software Stuff ### (Design, Architecture, Engineering & ...) === ### Reliability - Apps should continue to **work correctly** at the desired level of performance, even in the face of _adversity_. - Performance due to user expectation under the expected load and data volume - Tolerate user mistakes - Preventing any unauthorized access and abuse - _Fault-tolerant_ or _resilient_ systems anticipate faults and cope with them --- ### Hardware Faults - MTTF (mean time to failure) factor for hard disks - Add redundancy to the individual hardware components - More computing requirement => more machines => more hardware faults - Cloud platforms are designed to prioritize flexibility and elasticity over single-machine reliability - Using software fault-tolerance techniques in preference or in addition to hardware redundancy --- ### Human Errors - A Study: configuration errors by operators were the leading cause of outages, whereas hardware faults (servers or network) played a role in only 10–25% of outages --- ### Human Errors - Solutions - Well-designed abstractions, APIs, and admin interfaces - Provide fully featured non-production _sandbox_ environments - Test thoroughly at all levels - Quick and easy recovery from human errors - Roll back configuration changes - Roll out new code gradually - Tools to recompute data - Set up detailed and clear monitoring (_telemetry_) === ### Scalability - Dealing with system growth in data volume, traffic volume, or complexity - Define _load parameters_ based on system's architecture - Twitter scaling challenge: tweets _fan-out_ - sending posted tweets to timeline of the followers --- #### Performance Metrics (1/2) Gather performance metrics (_throughput_ or _response time_), calculate _percentiles_ - _Median_ (_p50_) - shows half of results - _p95_, _p99_, and _p999_, known as _tail latencies_ - Customers with the slowest requests are often those with the most data on their accounts, having made many purchases, and so most valuable ones - Amazon has observed a 100 ms increase in response time reduces sales by 1% - A 1-second slowdown reduces a customer satisfaction metric by 16% --- #### Performance Metrics (2/2) - Sample SLA - Median response time of less than 200 ms and a 99th percentile under 1 s - Uptime is at least 99.9% of the time - Measure response times on the client side - It prevents _head-of-line blocking_ miscalculation (in case of queuing requests, the longest ones result in total delay) === ### Maintainability - Different people do & will work _productively_. - 3 design principles for software systems - Operability - Make it easy for operations teams to keep the system running smoothly. - Simplicity - Make it easy for new engineers to understand the system. - Evolvability (_extensibility_, _modifiability_, or _plasticity_) - Make it easy for engineers to make changes to the system in the future. === ### References - Designing Data-Intensive Applications, Martin Kleppmann, OReilly, 2017