For those running an Network File System "NFS" server, the upcoming Linux 6.16 kernel will allow optionally enabling a larger payload size that may yield better performance. Eventually the default payload size may be increased if all goes well from user feedback and testing.

Chuck Lever of Oracle explained in the NFSD pull request for Linux 6.16:

"The marquee feature for this release is that the limit on the maximum rsize and wsize has been raised to 4MB. The default remains at 1MB, but risk-seeking administrators now have the ability to try larger I/O sizes with NFS clients that support them. Eventually the default setting will be increased when we have confidence that this change will not have negative impact.

With v6.16, NFSD now has its own debugfs file system where we can add experimental features and make them available outside of our development community without impacting production deployments. The first experimental setting added is one that makes all NFS READ operations use vfs_iter_read() instead of the NFSD splice actor. The plan is to eventually retire the splice actor, as that will enable a number of new capabilities such as the use of struct bio_vec from the top to the bottom of the NFSD stack.

Jeff Layton contributed a number of observability improvements. The use of dprintk() in a number of high-traffic code paths has been replaced with static trace points.

This release sees the continuation of efforts to harden the NFSv4.2 COPY operation. Soon, the restriction on async COPY operations can be lifted."

Those interested in upping the payload size for larger I/O can write their desired value in /proc/fs/nfsd/max_block_size and then restarting the NFS server so the larger payload/block size can be applied. Again, the default block size for the NFS server remains at 1MB but can be increased up to 4MB.