Simplified and Consolidated Parallel Media File System Solution

Author(s)/Presenter(s):
Library Content Type:
Publish Date: 
Thursday, September 27, 2018
Event Name: 
Focus Areas:
Abstract: 

Processing both metadata controller (MDC) I/O and file system data I/O workload on the same storage array reduces expense of equipment, rack space, and support costs for the customer by using under-utilized resources on the storage array to additionally service the metadata workload. This is accomplished by placing the metadata controller workload into a container on the same storage array that is processing the data workload for the file system and providing access to the data via ethernet bridging on the storage array.

We can process the data access to SCSI LUNs via the existing FC network (requiring no updates to the customer’s existing SAN infrastructure) for its high bandwidth workload and process the MDC I/O leveraging the NVMe coupling driver via the ethernet bridge for its periodic latency-sensitive workload. A common problem in existing M&E environments is complicated host clusters, which we have several proposals for simplification to build a robust solution with transient hosts. Finally, we propose a method to fence NVMe connections from SCSI task management functions which allows us to handle NVMe and SCSI on the same storage controller.

Learning Objectives:
1. How to put metadata controller workload on the same storage array as the data (client can access MDC via ethernet bridge)
2. How to share access with SCSI-based physical hosts and NVMe-based internal hosts via coupling driver (SCSI/NVMe inter-mix to support legacy network infrastructure)
3. How to provide easy host connection (volatile logins, multi LUN mapping, default host type, future: LUN groups, connectivity reporting/target failback into M&E, legacy rdac detect)
4. How to fence NVMe connections from SCSI Task Management Functions

Watch video: