NoSQL solutions cope with formerly unimaginable web-scale demands that are now a part of daily life. The term refers to data systems that relax the ACID restrictions (atomicity, consistency, isolation, and durability) that have long tied the hands of engineers, resulting in a gap as new user-facing web 2.0 applications have started to appear. In response, the big industry players have built flexible, elastic, and highly available platforms to respond to planetary-scale applications. This paper introduces Yahoo’s Platform for Nimble Universal Table Storage (PNUTS), a data-serving platform serving over 680 million customers from 18 global data centers. This system focuses on low-latency responses to satisfy users who will switch attention in less than half a second. PNUTS adheres to stringent service-level agreements (SLAs) regarding latency at the cost of less deterministic consistency.
The authors present the features of PNUTS in a technically shallow presentation. It is nevertheless a serious engineering work, empirically proven to be efficacious, with ingenious, straightforward solutions. PNUTS uses geographic asynchronous replication; timeline consistency; hash-based or ordered-based node mapping--and, consequently, simplified hardware scaling; a notification scheme similar to traditional database triggers; materialized views; and a versatile failure-recovery design, among other features. These traits save developers from having to worry about hardware provisioning, workload distribution, growth planning, and disaster contingency.
All in all, I recommend this paper. It is important to note that, although one can learn from it, the authors do not comparatively dissect the design and potential, and thus do not provide a basis for deciding which platform to use. For that kind of information, other sources are recommended [1].