Study and optimization of the memory management in Memcached

• Autor(a) principal: Coelho, Vasco Samuel Rodrigues
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/92295
• topic: Web cache; Memory management; Memcached; key-value stores; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Over the years the Internet has become more popular than ever and web applications like Facebook and Twitter are gaining more users. This results in generation of more and more data by the users which has to be efficiently managed, because access speed is an important factor nowadays, a user will not wait no more than three seconds for a web page to load before abandoning the site. In-memory key-value stores like Memcached and Redis are used to speed up web applications by speeding up access to the data by decreasing the number of accesses to the slower data storage’s. The first implementation of Memcached, in the LiveJournal’s website, showed that by using 28 instances of Memcached on ten unique hosts, caching the most popular 30GB of data can achieve a hit rate around 92%, reducing the number of accesses to the database and reducing the response time considerably. Not all objects in cache take the same time to recompute, so this research is going to study and present a new cost aware memory management that is easy to integrate in a key-value store, with this approach being implemented in Memcached. The new memory management and cache will give some priority to key-value pairs that take longer to be recomputed. Instead of replacing Memcached’s replacement structure and its policy, we simply add a new segment in each structure that is capable of storing the more costly key-value pairs. Apart from this new segment in each replacement structure, we created a new dynamic cost-aware rebalancing policy in Memcached, giving more memory to store more costly key-value pairs. With the implementations of our approaches, we were able to offer a prototype that can be used to research the cost on the caching systems performance. In addition, we were able to improve in certain scenarios the access latency of the user and the total recomputation cost of the key-value stored in the system.