Over at the RISElab blog, we have a post on the latest updates to the Anna KVS. Anna started out as the fastest KVS we are aware of (by orders of magnitude!), with the widest choice of consistency models. It achieved that in part by using beaucoup de resources to replicate entire databases in memory.
Thanks to recent work in the lab, Anna is not only incredibly fast, it’s incredibly efficient and elastic too: an autoscaling, multi-tier, selectively-replicating cloud service. All that adaptivity means that Anna ramps down resource consumption for cold things, and ramps up consumption for hot things. You get all the multicore Anna performance you want, but you don’t pay for what you don’t need.
Just to throw out some numbers, we measured Anna providing 355x the performance of DynamoDB for the dollar. No, I don’t think that is because AWS is earning a 355x margin on DynamoDB! The issue is that Anna is now orders of magnitude more efficient than competing systems, in addition to being orders of magnitude faster.
Chenggang Wu and Vikram Sreekanti were the driving forces behind this most recent work. They have more to say in their blog post, in the research paper on arXiv, and in the code on github.
(PS: the Anna v0 paper at ICDE was honored with a Best of Conference selection for publication in longer form in IEEE TKDE. Congrats to Chenggang Wu on that!)
Data in Beta 