Performance Tuning Tips and FAQs

This could be due to StorNext named streams being disabled by default on the system. See Enable StorNext named streams to enable named streams on your system.

NFS shares on managed filesystems require dmnfsthreads
to be set for shares that are StorNext Storage Manager relation points. If you see an error message similar to like "hostname:filesystem server not responding" on your NFS clients, "dmnfsthreads=" should be added to the StorNext filesystem mount option in the /etc/fstab
file on all NAS servers. If you have further questions about cvfs mount options, see man mount_cvfs
from the command line of the system.

If the Appliance Controller is not running on your nodes after a StorNext platform upgrade completes, it may be disabled. If this is the case, delete the etc/init/snnas_controller.override file and restart the Appliance Controller.

- Log in to the Appliance Controller CLI.
- Enter:
nascluster show
If the following error is returned, the Appliance Controller has been disabled:
Controller not running error: request to POST failed
- At the root shell prompt, enter the following command to verify that the Appliance Controller has been disabled by the etc/init/snnas_controller.override file:
# ls /etc/init/snnas_controller.override
If the /etc/init/snnas_controller.override file is returned, delete it. See Resolution.

- Log in to the command line using the following credentials:
- User name: stornext
- Password: <StorNext user account password>
Note: password is the default password for the stornext user account. If the password has been changed, use the current password. The first time you log in, you are prompted to change the password to a different one.
- Enter
sudo rootsh
to gain root user access. - Enter the password for the
stornext
user account again. - Enter the following to delete the etc/init/snnas_controller.override file:
# del /etc/init/snnas_controller.override
- At the prompt, enter the following to restart the Appliance Controller:
# initctl start snnas_controller

The default StorNext file system mount options are not optimized for Appliance Controller. If Appliance Controller performance is not optimal, you may need to tune StorNext to optimize performance.
We recommend tuning the following.

The cache buffer size is the amount of data that will be processed as a single block from the StorNext file system.
- Recommended setting for sequential I/O: See the mount_cvfs and cvtune man pages.
- Recommended setting for random I/O (such as database applications): typical read/write size
Example
If your typical read/write size is small (less than 32 KB) and random, set the cache buffer size to 64 KB.
-
Default setting: See the mount_cvfs and cvtune man pages.
Important
- Keep in mind that when you set the cache buffer size value, the StorNext file system will read/write the entire buffer size if the I/O is within EOF. If the workflow is writing I/O within a file that is only 4 KB and does not write the rest of the "cachebufsize", the system must read-modify-write the entire cachebufsize data block.
- When you increase the cache buffer size, you should also consider the buffer cache cap. See buffercachecap.
- You must optimize the raid array performance. See The Underlying Storage System in the StorNext 7 Documentation Center.

The buffer cache cap is the total amount of memory reserved for caching data. The reserved cache memory is shared by all mount points with the same cache size.
- Recommended Setting (dependent upon the amount of available memory): See the mount_cvfs and cvtune man pages.
- Default Setting: See the mount_cvfs and cvtune man pages.
When reserving cache memory, you need to take into account all processes that run on your system, such as smbd and FSM, so that you don't oversubscribe your system.

The directory cache size sets the size of the directory information cache on the Appliance Controller System. By increasing this value, the Appliance Controller System is able to keep more directory structure data in memory, dramatically improving the speed of readdir
operations by reducing metadata network message traffic between it and FSM.
- Recommend Setting: 32 MB
- Default Setting: 10 MB

The buffer cache I/O daemons setting defines the number of background daemons used for performing buffer cache I/O.
- Recommended Setting: 16
- Default Setting: 8
Example Command
$ snfs1 /stornext/snfs1 cvfs
$ rw,cachebufsize=256k,buffercachecap=4096,dircachesize=32m,buffercache_iods=16 0 0
For additional information, see the StorNextTuning and the StorNext 7 Man Pages Reference Guide.

We recommend modifying the following settings on your ProTools gateways for the best performance.

For best performance, consider modifying the following client settings:
Record buffer size
In ProTools 11, the default record buffer is around 25 seconds. We recommend increasing the record buffer if multiple tracks are being written together.
RAM
Ensure that the client system has sufficient RAM to configure Disk Cache for playback:
OS Memory + Playback Cache + Extra GB of Insurance

For best performance, consider modifying the following StorNext settings:
ASR
With high stream counts, we recommend setting Allocation Session Reservation (ASR) to 256 MB to keep the streams from interleaving. The default setting of 1 GB is overkill for audio settings.
For more information about ASR, see Allocation Session Reservation (ASR) in the StorNext 7 Documentation Center.
Client Buffer Cache
We recommend increasing client buffer cache settings to 1GB or more to smooth record and play back.
Buffer Cache IODs
For high track count environments, we recommend that you add spindles to help with Buffer Cache IODs. The additional spindles help to address the potential limiting factor of disk actuators as the disks are seeking across multiple tracks.
Example
$ snfs1 /stornext/snfs1 cvfs
$ rw, threads=16,cachebufsize=256k,buffercachecap=4096,dircachesize=128m,buffercache_iods=16 0 0

Note: This issue applies to Mac OS X 10.11.5+ and Mac OS Sierra.
Apple's forced default of SMB client signing can cause significant decreases in SMB file transfer times on your Mac OS X 10.11.5+ or Mac OS Sierra. To resolve this issue, disable the forced client-signing on all necessary Mac clients.

- From the Mac client, access the /etc/nsmb.conf file.
- Populate this file with the following lines:
[default]
signing_required=no
- Unmount and remount the SMB share.
SMB file transfer times should return to normal.

When using the Mac Finder to look at a directory on an SMB share, Finder attempts to individually retrieve every file to create preview of it. If many files are truncated, this process can cause the SMB client to timeout. To prevent the SMB client from retrieving truncated files for a certain share, or globally:
- Log in to the Appliance Controller CLI.
- Do one of the following:
-
To prevent the SMB client from retrieving truncated files for a certain share, enter:
Xcellis:xcellis1>
share change smb smb1 snfs:allow retrieve = no
Output:
Share smb1 successfully changed
-
To prevent the SMB client from globally retrieving truncated files, enter:
Xcellis:xcellis1>
share change smb global snfs:allow retrieve = no
Output:
Share global successfully changed
- Exit the Appliance Controller.