Isilon: OneFS-How to enable access time tracking (atime)
Summary: This article explains how to enable access time tracking (atime) for OneFS. By default atime is not enabled.
Instructions
Introduction
Access time, known by the shortened form, "atime," is the time at which a file was last accessed. Access time tracking records the time at which files were created and last modified as well as when files were last accessed.
In OneFS, full access time tracking is supported but disabled by default. However, you can enable access time tracking if necessary.
For example, you must enable access time tracking if you want to configure SyncIQ policy criteria that match files based on when they were last accessed.
This article provides procedures on how to enable the atime feature using the OneFS web administration interface or the command line.
CAUTION!
Updating access time tracking adds a write operation to every access operation. As such, enabling access time tracking can have a significant negative effect on cluster performance. Also, in Server Message Block (SMB) environments, a change notify message is sent with every access operation. This can cause heavy network traffic and prevent SMB clients from navigating shares on the cluster.
It is recommended that you enable this option only if necessary.
Procedure
Using the OneFS web administration interface
- Click File System Management > File System Settings > Access Time Tracking.
- Select the Access time tracking Enabled option.
- Edit the Precision options as wanted to specify (in Seconds, Minutes, Hours, Days, Weeks, Months, or Years) how often to update last-accessed times. For example, if you were to configure a Precision setting of one Day, the cluster would update the last-accessed time once per day, even if some files were accessed more often than once per day.
IMPORTANT!
Do not set to precision to less than one hour. As a best practice, set precision to at least one day.
- Click Submit.
NOTE: Access Time only applies going forward, not to files previously accessed prior to when access time was enabled for the mount.
Using the command-line interface
- Open an SSH connection on any node in the cluster and log in using the "root" account.
- Run the following command:
# isi_for_array -s sysctl efs.bam.atime_enabled=1 Confirm change has been applied to the cluster: isilon-2# isi_for_array -s sysctl efs.bam.atime_enabled isilon-1: efs.bam.atime_enabled: 1 isilon-2: efs.bam.atime_enabled: 1 isilon-3: efs.bam.atime_enabled: 1
NOTE
The above command enables atime, but it is not permanent. Any node or cluster reboot can revert and disable the setting. - To enable access time tracking permanently, add it to the system override file by running the following command:
CAUTION!
Improper implementation of sysctl commands can render your cluster inoperable.# isi_sysctl_cluster efs.bam.atime_enabled=1 Confirm change has been applied to the cluster: isilon-2# isi_for_array -s sysctl efs.bam.atime_enabled isilon-1: efs.bam.atime_enabled: 1 isilon-2: efs.bam.atime_enabled: 1 isilon-3: efs.bam.atime_enabled: 1
NOTE
This creates a /etc/mcp/override/sysctl.conf file if it did not previously exist.
Setting the precision using the command line
The
atime_grace_period Setting on the command line corresponds with the "Precision" setting in the OneFS web administration interface. It governs the minimal amount of time that must pass between file accesses before OneFS updates the last-accessed time. Higher settings reduce the accuracy of the setting but also mitigate the performance penalty associated with recording atime on a file access.
The setting is in milliseconds, and the default setting is:
86400000 (1 day)
- Run the following command and confirm the change:
# isi_for_array -s sysctl efs.bam.atime_grace_period=<number> Confirm change has been applied to the cluster: # isi_for_array -s sysctl efs.bam.atime_grace_period
IMPORTANT!
Do not set precision to less than one hour (3600000). As a best practice, set precision to at least one day ( 86400000). - To change the default grace period permanently, run the following command:
CAUTION!
Improper implementation of sysctl commands can render your cluster inoperable and reducing the atime to less than one hour (3600000) is not advised. Severe performance impact can occur when set to less than one hour.# isi_sysctl_cluster efs.bam.atime_grace_period=86400000 Confirm change has been applied to the cluster: # isi_for_array -s sysctl efs.bam.atime_grace_period
After you have enabled access time tracking, the system writes the access-time tracking information, with the current time when the file is accessed. The following example shows access time tracking for file1. It has a last access time of almost two months prior to the initial timestamp, and the cluster is set with a precision of 1 hour (
3600000). The info is written to the metadata of the file. You can run #isi get -DDd <file name> before enabling the feature, and after enabling the feature. The last access entry is added to the output.
Initial Timestamp of file1: 1233383074 01/31/2009 @ 12:24am 1236931800 03/13/2009 @ 3:10 - File is opened: Outside of grace period. file1 atime is updated to 1236931800 1236932800 03/13/2009 @ 3:26 - File is opened again: Within grace period. file1 atime stays at 1236931800 1236939000 03/13/2009 @ 5:10 - File is opened again: Outside of grace period. file1 atime is updated to 1236939000