I have two Linux systems: NFSServer1 (RHEL) and NFSClient1 (Ubuntu).
On NFSServer1, ntfs-3g driver and ldmtool is installed. The NTFS device partitions are mounted by executing the command:
mount -t ntfs-3g -o ro,noatime $devPath $mountPath
Note: The two partitions /dev/mapper/ldm_vol_VishalWDD-Dg0_Volume1 and /dev/mapper/ldm_vol_VishalWDD-Dg0_Volume2 are Windows Dynamic disk partitions derived using ldmtool
[root@ROADQAScaleNFS2 ~]# df -Th
Filesystem Type Size Used Avail Use% Mounted on
/dev/sdc4 fuseblk 127G 11G 117G 9% /monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/sdc4
/dev/sdc2 fuseblk 450M 13M 438M 3% /monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/sdc2
/dev/mapper/ldm_vol_VishalWDD-Dg0_Volume1 fuseblk 10G 5.8G 4.3G 58% /monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/mapper/ldm_vol_VishalWDD-Dg0_Volume1
/dev/mapper/ldm_vol_VishalWDD-Dg0_Volume2 fuseblk 10G 6.0G 4.1G 60% /monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/mapper/ldm_vol_VishalWDD-Dg0_Volume2
[root@ROADQAScaleNFS2 ~]# mount
/dev/sdc4 on /monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/sdc4 type fuseblk (ro,noatime,user_id=0,group_id=0,allow_other,blksize=4096)
/dev/sdc2 on /monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/sdc2 type fuseblk (ro,noatime,user_id=0,group_id=0,allow_other,blksize=4096)
/dev/mapper/ldm_vol_VishalWDD-Dg0_Volume1 on /monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/mapper/ldm_vol_VishalWDD-Dg0_Volume1 type fuseblk (ro,noatime,user_id=0,group_id=0,allow_other,blksize=4096)
/dev/mapper/ldm_vol_VishalWDD-Dg0_Volume2 on /monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/mapper/ldm_vol_VishalWDD-Dg0_Volume2 type fuseblk (ro,noatime,user_id=0,group_id=0,allow_other,blksize=4096)
Totally all these partitions have about 10 million files.
These mounted partitions are accessed from NFSClient1 as NFS shares:
[root@NFSClient ~]# mount
10.4.0.5:/monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/sdc4 on /monitor1/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/sdc4 type nfs4 (ro,relatime,vers=4.2,rsize=1048576,wsize=1048576,namlen=255,hard,proto=tcp,timeo=600,retrans=2,sec=sys,clientaddr=10.148.66.49,local_lock=none,addr=10.4.0.5)
10.4.0.5:/monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/sdc2 on /monitor1/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/sdc2 type nfs4 (ro,relatime,vers=4.2,rsize=1048576,wsize=1048576,namlen=255,hard,proto=tcp,timeo=600,retrans=2,sec=sys,clientaddr=10.148.66.49,local_lock=none,addr=10.4.0.5)
10.4.0.5:/monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/mapper/ldm_vol_VishalWDD-Dg0_Volume1 on /monitor1/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/mapper/ldm_vol_VishalWDD-Dg0_Volume1 type nfs4 (ro,relatime,vers=4.2,rsize=1048576,wsize=1048576,namlen=255,hard,proto=tcp,timeo=600,retrans=2,sec=sys,clientaddr=10.148.66.49,local_lock=none,addr=10.4.0.5)
10.4.0.5:/monitor/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/mapper/ldm_vol_VishalWDD-Dg0_Volume2 on /monitor1/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/mapper/ldm_vol_VishalWDD-Dg0_Volume2 type nfs4 (ro,relatime,vers=4.2,rsize=1048576,wsize=1048576,namlen=255,hard,proto=tcp,timeo=600,retrans=2,sec=sys,clientaddr=10.148.66.49,local_lock=none,addr=10.4.0.5)
The number of NFS daemon threads on NFS server is set to 64.
Next, on the NFS client, when we issue a stat a fuseblk partition using a find command:
find -H /monitor1/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/mapper/ldm_vol_VishalWDD-Dg0_Volume2 -printf '%p|' | xargs -d '|' stat --printf="%F, %i:\t%n\t%.19x\t%.19y\t%.19z\t%.19w\t%s\t%u\t%g\n" \ |$SED -e "s|/monitor1/6f5bd42-e548-4e60-8c5d-4c52360b8dc4/mapper/ldm_vol_VishalWDD-Dg0_Volume2/||g" -e "s|directory,|d/d|g" -e "s|symbolic link,|l/l|g" -e "s|regular file,|r/r|g" -e "s|socket,|h/h|g" \ -e "s|regular empty file,|r/r|g" -e "s|fifo,|p/p|g"
It execution is extremely slow. It takes a break of 5-6 minutes and then resumes for a few seconds. The same is true for all other mount points. The execution does not finish even after 12 hours.
This sluggish behavior is not observed for ext4 and xfs devices types.
As a test, I tried executing the same find command on the NFSServer1, it was quite fast. The whole execution finished in ~40 minutes. I don't have access to the NFS server though. I have asked the NFS server team to try different mount options as mentioned in the ntfs-3g man page, but it didn't help.
If there is any way to improve the read performance of fuseblk partitions over NFS, I would grateful to you guys.
Many thanks!
