NFSv4 on B3, where is the bottleneck (writing to B3)

[ingo2]

I have set up the nfs-kernel-server on my B3.
Created a separate "sever-partition" sda4 which I mount to /srv/sda4 in fstab. Then I created the nfs4 mount-tree under /srv/nfs4 like this:

Code: Select all

/etc/fstab
# Partititon /dev/sda4 (HD with GPT) for servers
/dev/sda4	/srv/sda4	ext4	defaults,noatime	0  3
# Bindings for NFS v4 below /srv/nfs4
/srv/sda4/backup        /srv/nfs4/backup	none	bind	0  0

Code: Select all

/etc/exports
# Exports for nfs v4
/srv/nfs4        192.168.0.0/24(fsid=0,rw,insecure,sync,no_subtree_check)
/srv/nfs4/backup 192.168.0.0/24(nohide,rw,insecure,sync,no_subtree_check,no_root_squash)

I usually only mount the "backup" export on my PC and observed the following issue:

Reading from the B3 works great with 45MB/s, both many small and also large files.
Writing to the B3 however (with large files) starts at ~35MB/s during first 750MB and then slows down to kreeping 1MB/s.

Top shows almost no CPU-load, but 97% wa(iting for I/O).

To solve this for the moment I had to modify export options on B3 from "sync" -> "async".
This way I can write continuosly at 32MB/s independent of file size. However for data integrity this is not the best choice.

Does anybody know where the bottleneck for writing is?
I already checked for swapping, but swap is not used at all while writing rate is so slow. Also the HD itself can be excluded, as rate is fine with "asysnc" writing.

With best regards,
Ingo