Self-learning Caches

Salon B/C/D

Caches in modern storage systems are static and unable to learn and adapt. There is strong evidence that the application of self-learning into storage caches can unlock huge improvements in cost, efficiency, performance and predictability. Cache are inherently nonlinear systems making their tuning an exercise equivalent to solving a maze in the dark. This is precisely why the application of self-learning techniques can yield such a huge improvement against the static baseline.
In this talk, we will begin by providing an overview of the opportunity that self-learning caches promise by looking at some case studies. Irrespective of whether the cache is in a storage system, database or application tier, no two real workload mixes have the same cache behavior so we will look at several different real examples. Based on research recently published (USENIX ATC '17), we will review techniques for online, real-time self-learning caches and describe how they achieve huge gains.

Learning Objectives


  • Why static caches leave so much performance on the floor
  • What is a self-learning cache and how does it automatically adapt to new workloads
  • The efficiency, QoS, performance SLAs/SLOs and cost tradeoffs that self-learning caches enable
  • The broader application of self-learning cache technology to multi-tier and multi-tenant systems


Cache Physics
Irfan Ahmad is the CEO and Cofounder of CachePhysics. Previously, Irfan served as the CTO of CloudPhysics which he cofounded in 2011. Under his leadership, CloudPhysics brought to market and scaled the industry's first SaaS Virtualized IT Operations Management product. Irfan was at VMware for 9 years, where he was R&D tech lead for the DRS team and co-inventor for flagship products, including Storage DRS and Storage I/O Control. Before VMware, Irfan worked on the Crusoe software microprocessor at Transmeta. Irfan is an inventor on more than twenty five issued and ten pending patents. Irfan has published at ACM SOCC, USENIX FAST, ATC and IEEE IISWC and has chaired HotStorage, HotCloud and VMware’s R&D Innovation Conference. Irfan serves on steering committees for USENIX HotStorage and HotCloud. Irfan has served on program committees for FAST, MSST, HotCloud and HotStorage among others and as a reviewer for the ACM Transactions on Storage. Irfan earned his pink tie from the University of Waterloo.