Workload analysis of a large-scale key-value store

Abstract

Key-value stores are a vital component in many scale-out enterprises, including social networks, online retail, and risk analysis. Accordingly, they are receiving increased attention from the research community in an effort to improve their performance, scalability, reliability, cost, and power consumption. To be effective, such efforts require a detailed understanding of realistic key-value workloads. And yet little is known about these workloads outside of the companies that operate them. This paper aims to address this gap.To this end, we have collected detailed traces from Facebook's Memcached deployment, arguably the world's largest. The traces capture over 284 billion requests from five different Memcached use cases over several days. We analyze the workloads from multiple angles, including: request composition, size, and rate; cache efficacy; temporal patterns; and application use cases. We also propose a simple model of the most representative trace to enable the generation of more realistic synthetic workloads by the community.Our analysis details many characteristics of the caching workload. It also reveals a number of surprises: a GET/SET ratio of 30:1 that is higher than assumed in the literature; some applications of Memcached behave more like persistent storage than a cache; strong locality metrics, such as keys accessed many millions of times a day, do not always suffice for a high hit rate; and there is still room for efficiency and hit rate improvements in Memcached's implementation. Toward the last point, we make several suggestions that address the exposed deficiencies.

Workload Modeling for Computer Systems Performance Evaluation

“…Characterization of stored data in Facebook servers, which is organized as a key-value store, revealed that both keys and values tend to have modal distributions [42]. File sizes also tend to be modal.…”

“…• Accesses to the records of a database display locality, as do accesses to a key-value store, which may be regarded as a simple case of a database [42].…”

“…The diurnal cycle is a well-known and widely recognized phenomenon, which appears in practically all workloads (e.g., [326,320,524,63,761,745,756,381,192,42]). An example from parallel supercomputers is shown in Figure 6.34.…”

“…If this is insufficient, a finer resolution of one hour can be used [539,42]. The other approach is to use a parameterized distribution, and to model the daily cycle by showing how the parameters change with the time of day.…”

“…In addition, note that MapReduce is defined in terms of key-value pairs, and therefore the handling and caching of such data is of great importance. Atikoglu et al report on the workload handled by five large key-value stores at Facebook [42]. Their main findings are as follows.…”