B Mount Options in a Shared File System

An Oracle Hierarchical Storage Manager and StorageTek QFS Software shared file system can be mounted with several mount options. This chapter describes some of these options within the context of their roles.

Shared File System Mount Options

You can specify most mount options by using the mount command, by entering them in the /etc/vfstab file, or by entering them in the samfs.cmd file. For example, the following /etc/vfstab file includes mount options for a shared file system:

sharefs - /sfs samfs - no shared,mh_write

You can change some mount options dynamically by using the samu operator utility. For more information about these options, see the Oracle Hierarchical Storage Manager and StorageTek QFS samuCommand Reference.

For more information about any of these mount options, see the mount_samfs man page.

bg: Mounting in the Background

The bg mount option specifies that if the first mount operation fails, subsequent attempts at mounting should occur in the background. By default, bg is not in effect, and mount attempts continue in the foreground.

retry: Reattempting a File System Mount

The retry mount option specifies the number of times that the system should attempt to mount a file system. The default is 10000.

shared: Declaring a Oracle HSM Shared File System

The shared mount option declares a file system to be an Oracle HSM shared file system. This option must be specified in the /etc/vfstab file in order for the file system to be mounted as an Oracle HSM shared file system. The presence of this option in a samfs.cmd file or on the mount command does not cause an error condition, but it does not mount the file system as a shared file system.

minallocsz and maxallocsz: Tuning Allocation Sizes

The minallocsz and maxallocsz options to the mount command specify an amount of space, in kilobytes. These options set the minimum block allocation size. If a file is growing, the metadata server allocates blocks when an append lease is granted. Use -o minallocsz=n to specify the initial size of this allocation. The metadata server can increase the size of the block allocation depending on the application's access patterns up to but not exceeding the -o maxallocsz=n setting.

You can specify these mount options on the mount command line, in the /etc/vfstab file, or in the samfs.cmd file.

rdlease, wrlease, and aplease: Using Leases in a Oracle HSM Shared File System

By default, when hosts share files, the Oracle HSM metadata server maintains file-system consistency by issuing I/O leases to itself and its clients. A lease grants a shared host permission to perform an operation on a file for a specified period. A read lease lets a host read file data. A write lease lets a host overwrite existing file data. An append lease lets a host write additional data at the end of a file. The metadata server can renew leases as necessary.

Reads and writes to a Oracle HSM shared file system should thus provide near-POSIX behavior for data. For metadata, however, access time changes might not be seen immediately on other hosts. Changes to a file are pushed to disk at the end of a write lease. When a read lease is acquired, the system invalidates any stale cache pages so that the newly written data can be seen.

The following mount options set the duration of the leases:

• -o rdlease= number-seconds specifies the maximum amount of time, in seconds, for the read lease.

• -o wrlease= number-seconds specifies the maximum amount of time, in seconds, for the write lease.

• -o aplease= number-seconds specifies the maximum amount of time, in seconds, for the append lease.

In all three cases, number-seconds is an integer in the range [15-600]. The default time for each lease is 30 seconds. A file cannot be truncated if a lease is in effect. For more information about setting these leases, see the mount_samfs man page.

If you change the metadata server because the current metadata server is down, you must add the lease time to the changeover time because all leases must expire before an alternate metadata server can assume control.

Setting a short lease time causes more traffic between the client hosts and the metadata server, because the lease must be renewed after it has expired.

mh_write: Enabling Multiple Host Reads and Writes

The mh_write option controls write access to the same file from multiple hosts. If mh_write is specified as a mount option on the metadata server host, the Oracle HSM shared file system enables simultaneous reads and writes to the same file from multiple hosts. If mh_write is not specified on the metadata server host, only one host can write to a file at any one time.

By default, mh_write is disabled, and only one host has write access to a file for the duration of the wrlease mount option. If the Oracle HSM shared file system is mounted on the metadata server with the mh_write option enabled, simultaneous reads and writes to the same file can occur from multiple hosts.

When mh_write is enabled on the metadata server, Oracle HSM supports the following:

• Multiple reader hosts and paged I/O

• Multiple reader and/or writer hosts and direct I/O only if there are writers

• One append host (other hosts read or write) and direct I/O only if there are writers.

Mounting a file system with the mh_write option does not change locking behavior. File locks behave the same regardless of whether mh_write is in effect. But, in other respects, behavior might be less consistent. When there are simultaneous readers and writers, the Oracle HSM shared file system uses direct I/O for all host access to a file. Therefore, page-aligned I/O should be visible immediately to other hosts. However, non-page-aligned I/O can result in stale data being visible, or even written to the file, because the normal lease mechanism preventing such occurrences has been disabled.

For this reason, you should specify the mh_write option only when multiple hosts need to write the same file simultaneously and when hosted applications perform page-aligned I/O and coordinate conflicting writes. In other cases, data inconsistency could occur. Using flock() with mh_write to coordinate between hosts does not guarantee consistency. For more information, see the mount_samfs man page.

min_pool: Setting the Minimum Number of Concurrent Threads

The min_pool mount option sets the minimum number of concurrent threads for the Oracle HSM shared file system. The default setting is min_pool=64 on Oracle Solaris systems. This setting means that at least 64 active threads will be in the thread pool on Oracle Solaris. You can adjust the min_pool setting to any value in the range [8-2048], depending on shared file-system activity.

The min_pool mount option must be set in the samfs.cmd file. It will be ignored if set in the /etc/vfstab file or at the command line.

meta_timeo: Retaining Cached Attributes

The meta_timeo mount option determines how long the system waits between checks on the metadata information. By default, the system refreshes metadata information every three seconds. For example, an ls command entered in a shared file system with several newly created files might not return information about all the files until three seconds have passed. The syntax for the option is meta_timeo=seconds where seconds is an integer in the range [0-60].

stripe: Specifying Striped Allocation

By default, data files in a shared file system are allocated using the round-robin file allocation method. To specify that file data be striped across disks, you can specify the stripe mount option on the metadata host and all potential metadata hosts. Note that by default, unshared file systems allocate file data using the striped method.

In a round-robin allocation, files are created in a round-robin fashion on each slice or striped group. The maximum performance for one file will be the speed of a slice or striped group. For more information about file allocation methods, see the Oracle Hierarchical Storage Manager and StorageTek QFS Installation and Configuration Guide (Oracle HSM Customer Documentation Library, docs.oracle.com/en/storage).

sync_meta: Specifying the Frequency With Which Metadata Is Written

You can set the sync_meta option to sync_meta=1 or sync_meta=0.

The default setting is sync_meta=1, which means that a Oracle HSM shared file system writes file metadata to disk every time the metadata changes. This setting slows data performance but ensures data consistency. This setting must be in effect if you want to change the metadata server.

If you set sync_meta=0, the Oracle HSM shared file system writes the metadata to a buffer before writing it to disk. This delayed write delivers higher performance but decreases data consistency after an unscheduled machine interruption.

worm_capable and def_retention: Enabling WORM Functionality

The worm_capable mount option lets the file system support WORM files. The def_retention mount option sets the default retention time using the format def_retention=MyNdOhPm.

In this format, M, N, O, and P are non-negative integers and y, d, h, and m stand for years, days, hours, and minutes, respectively. Any combination of these units can be used. For example, 1y5d4h3m indicates 1 year, 5 days, 4 hours, and 3 minutes; 30d8h indicates 30 days and 8 hours; and 300m indicates 300 minutes. This format is backward compatible with the formula in previous software versions, in which the retention period was specified in minutes.

For more information, see the Oracle Hierarchical Storage Manager and StorageTek QFS Installation and Configuration Guide (Oracle HSM Customer Documentation Library, docs.oracle.com/en/storage).