Responsible for the development of in kernel metadata consistency checker for Vmware's VSAN hyper-converged stack. Worked on the change block tracking system which has an in band metadata mechanism for dealing with consistency of replicas as well as tracking changes from a given consistency point.

Implemented a IO scheduling scheme for shaping and regulating disk IOs below the ZFS vdev layer to cater to multi-tenant qos requirements. GEOM layer bottleneck removal, context switches and queue contentions etc. Design approaches for ZFS metadata acceleration with optimal placement and preferential write allocations, persistent caches with recovery and rebuild support for reboot or disk failure events. ZIL fast path preferential scheduling and IO prioritization. Freebsd IO/CPU performance bottleneck identification and removal with hwpmc,dtrace,lock profiling, sleep queue and scheduler profiling etc. Implemented an ISCSI parallel processing scheme which is adaptive to go parallel or serial so as to safely account for write after read, read after write,control/reset dependencies among concurrent requests.

Responsible for the architecture and implementation of a Big Data Content repository (several PB with billions of files/objects) which can be accessed through NAS protocols (NFS and CIFS) as well as the Cloud Data Management Interface (CDMI). Designed critical features for posix compliance, layout of the files/objects for supporting regular attributes and extended attributes like ACL,streams,custom metadata and associated cluster lock state distribution needed to support delegations,L1/L2 oplocks,delete on close. Implemented non disruptive movement of files/objects with intelligent redirection for capacity balancing. Design for exporting the file/object layout for high performance parallel access via pNFS. Prototyping the protocol engine designs for driving http workloads.

Pathless objects accessible via nfs IETF draft

http://tools.ietf.org/html/draft-dipankar-nfsv4-pathless-objects-02