Chapter 2: Back-of-the-Envelope Estimation

Proficiency in scalability fundamentals is necessary for accurate estimation, starting with a mastery of the power of two for correctly calculating the unit volume of data, which is crucial when dealing with modern distributed systems that handle enormous amounts of data ranging from kilobytes (KB) to petabytes (PB). A key focus involves analyzing the latency numbers associated with various computer operations to understand performance differences. These numbers clearly illustrate that memory access, such as an L1 cache reference (0.5 nanoseconds), is dramatically faster than operations involving physical storage, like a disk seek (10 milliseconds). Analyzing these performance differences leads to vital conclusions for optimization: memory is exceptionally fast while disk access is slow, emphasizing the need to avoid disk seeks whenever possible, and highlighting that simple data compression algorithms are fast and should be used before sending data over a network to reduce transfer time. The concept of high availability is also explored, representing a system’s ability to remain continuously operational, measured as a percentage, often governed by a Service Level Agreement (SLA) with cloud providers aiming for 99.9% or greater uptime. This availability is traditionally quantified in "nines," where additional nines correspond to significantly less expected downtime per year. The application of these principles is demonstrated through practical examples, such as estimating the Query Per Second (QPS) and substantial 5-year storage needs (potentially 55 PB) for a high-volume platform. Candidates are advised to write down all assumptions, label units clearly to eliminate ambiguity, and utilize rounding and approximation to simplify complex calculations, focusing on the quality of the problem-solving process rather than achieving absolute precision.