Talk Title: D4N

Talk Abstract: Shared data lakes are a premier setting for the convergence of data and computation, and home to critical workloads such as big data/analytics and AIML.  Red Hat/IBM and MOC have collaborated since 2017 on a series of research projects in this space, including workload acceleration through intelligent caching, and more recently computational storage.  Their most recent collaborations center on D4N, an intelligent caching framework integrated into Ceph’s object storage gateway.  Team members have focused on contributing these developments into Ceph’s open source community, using new flexible abstractions and writeback caching–including resilient storage converged with the object storage tier.  Within IBM, related work has focused on new computational storage access paths and abstractions (e.g., S3select, Arrow Flight).   We would like to understand the interest in the Open Cloud community for using D4N, and if the community would be interested in exploiting it as a best effort experimental service if we make it available before it becomes a supported product.

Bios:

Matt Benjamin is the engineering manager for the Ceph Object Storage and user-space NFS teams at IBM, and past PI for CohortFS,  an NFS-funded research project that explored parallel NFS access and  performance optimizations in the Ceph distributed storage system.

Amin Mosayyebzadeh is a Ph.D. candidate in the Department of Electrical and Computer Engineering at Boston University working with Prof. Orran Krieger and Prof. Peter Desnoyers. He is a member of the Systems Research Group and is associated with Mass Open Cloud (MOC). His research interests lie broadly in Distributed Systems, Storage Systems, and Cloud Computing. Currently, he is D4N, an intelligent distributed caching system.