Here are a few ongoing problem areas where I feel it would be useful to have a group conversation. These are not plenary topics by any means, but might be interesting as part of a BoF or break- out session for attendees interested in network file systems. - NFS performance Identify common goals, see if there's low-hanging fruit. There are common issues for Linux's remote filesystem implementations related to memory allocation, scheduler behavior, and network APIs that would be useful to discuss. How do we plan to leverage persistent and storage class memory? What kind of continuous performance engineering is going on? - Recovery after temporary or permanent NFS server outages We discussed this during last year's LSF but the conversation is not finished. Jeff's writeback error reporting work enables the NFS client to recover from server outages the same way that local filesystems might recover from a failed disk. We need the NFS client to behave better in these scenarios. Indeed, container environments need this kind of recovery to reduce friction between tenants that share NFS resources. - Kernel DNS service NFSv4 referrals can return a list of server hostnames, or hostnames that resolve to multiple IP addresses. Our remote filesystem client implementations handle only a single address in these cases, and that limits the ability to select the best network path to the server. -- Chuck Lever
