Hi all,
I have seen some behaviour that I need to confirm and need some advise please...
I have a PXE boot system running via NFS. This is working well, but have a question around mount points and performance of memory disks.
I have a number of memory disks that are created via cpio zip files in /conf/base/ of the NFS root, and are installed via /etc/rc.initdiskless when the PXE system boots. The system root (/) on the PXE boot machine is a read-only (RO) NFS mount via an entry in /etc/fstab. On the RO NFS there is a folder called /test/ which acts as a mount point for a memory disk.
The memory disk is successfully mounted at /test/, and I can read and write files without error.
The oddity I am seeing is that when accessing files within the /test/ folder, For each file that is accessed on the memory disk (mounted at the /test/ folder of the RO NFS), e.g. /test/some_file_name, an NFS v3 query for /test/ is sent from the PXE booted machine to the NFS server, of which the NFS responds to with an NFS_OK message. This is for every file in the memory disk! This means that the memory disk performance is subject to network jitter, latency and potential congestion.
Needless to say, this is of course having a wild impact on performance versus expected performance of the memory disk, as all access is preceded with a "slow" NFS query over the network.
Any ideas if the behaviour of NFS with regard to queries to a mount point can be improved?
Ideas or thoughts very much welcomed!
James.
I have seen some behaviour that I need to confirm and need some advise please...
I have a PXE boot system running via NFS. This is working well, but have a question around mount points and performance of memory disks.
I have a number of memory disks that are created via cpio zip files in /conf/base/ of the NFS root, and are installed via /etc/rc.initdiskless when the PXE system boots. The system root (/) on the PXE boot machine is a read-only (RO) NFS mount via an entry in /etc/fstab. On the RO NFS there is a folder called /test/ which acts as a mount point for a memory disk.
The memory disk is successfully mounted at /test/, and I can read and write files without error.
The oddity I am seeing is that when accessing files within the /test/ folder, For each file that is accessed on the memory disk (mounted at the /test/ folder of the RO NFS), e.g. /test/some_file_name, an NFS v3 query for /test/ is sent from the PXE booted machine to the NFS server, of which the NFS responds to with an NFS_OK message. This is for every file in the memory disk! This means that the memory disk performance is subject to network jitter, latency and potential congestion.
Needless to say, this is of course having a wild impact on performance versus expected performance of the memory disk, as all access is preceded with a "slow" NFS query over the network.
Any ideas if the behaviour of NFS with regard to queries to a mount point can be improved?
Ideas or thoughts very much welcomed!
James.