Memcached Architecture | The Backend Engineering Show

This episode explores the inner workings of Memcached, a popular in-memory key-value store. Against the backdrop of its simple design and historical context (originating in 2003 and used by companies like Facebook and Netflix), the podcast delves into its architecture. More significantly, the host discusses Memcached's memory management, highlighting the challenges of fragmentation and the employed solution of allocating pages and chunks to avoid this. The least recently used (LRU) algorithm is examined, with the host expressing a personal preference for disabling it by default due to its performance overhead. Further discussion covers the threading model, initially serialized but later improved to a per-item lock, and the hash table used for key lookups, including collision handling. Finally, the podcast clarifies that Memcached's distribution is client-side, not inherent to the server itself, emphasizing its design philosophy of simplicity. This means for developers, a deep understanding of Memcached's internal mechanisms allows for optimized application architecture and efficient key-value storage.