Skip Headers
Oracle® Hierarchical Storage Manager and StorageTek QFS Software samu User Interface Guide
Release 6.0
E48431-03
  Go To Documentation Library
Library
Go To Table Of Contents
Contents

Previous
Previous
 
Next
Next
 

2 Commands Listed by Function

This chapter groups the samu commands by task and function. Locate the task that you need to carry out and then see the alphabetical command listings in Chapter 3, "Operator Commands", and Chapter 4, "Service and Support Commands", for more information:

Tuning I/O Performance

This section starts with a brief summary of the factors that affect file-system performance, the objectives of performance tuning, and the Oracle HSM features that support tuning efforts. It then links to the commands that control the input/output methods that Oracle HSM supports: Paged (Buffered) I/O, Direct I/O, and Switched I/O.

Understanding I/O Performance Tuning Objectives and Options

Disk I/O (input/output) involves mechanical processes that are vastly more time-consuming than other file-system operations. So I/O performance tuning focuses on keeping the mechanical work to the absolute minimum necessary for transferring a given amount of data. This means reducing both the number of separate I/Os per data transfer and the number of seeks operations required to service each I/O.

So the basic objectives of I/O tuning are as follows:

  • Read and write large blocks of data.

  • Write blocks in units that align with the sector boundaries of the underlying media, so that the disk controller does not have to read and modify existing data before writing the new data.

  • Queue up small I/Os in cache and write larger, combined I/Os to disk.

To achieve these objectives, Oracle HSM file systems support three kinds of I/O, each of which addresses a different set of file-system usage patterns.

Paged (Buffered) I/O is the default method because it works well with the usage characteristics of common, general-purpose file-systems. Most file-systems are used by multiple users and applications. Each reads and writes small amounts of data in blocks that often depend on application design rather than on disk sector boundaries. File access tends to be random rather than sequential. When properly configured, paged I/O is ideal for this kind of use. User data is cached in virtual memory pages until the Oracle Solaris kernel writes the data to disk. So users and applications write to and read from cache memory, where random access and small block sizes are less of a problem. Performance-critical, physical reads and writes to and from disk are made in larger, better aligned, more nearly sequential chunks.

Direct I/O addresses limitations of paged I/O that become apparent in file systems are dedicated to particular types of data, such as medical imagery, geological information bases, and real-time surveillance imagery. Generally, users and applications read and write large files sequentially in blocks that align with disk sector boundaries. So paging has no performance advantages. It does, however, impose system overhead and delays the point at which data is securely written from memory to disk—the latter potentially critical in realtime applications. In this situation, direct I/O can significantly improve performance. Data is transferred directly between the local buffer on the application host and the Oracle HSM disk device, with no intermediate delays or additional complication.

Switched I/O is a combination of the other two that works by setting an I/O size threshold for the file system. Paged I/O is used until this threshold is reached. Thereafter, the system automatically switches to direct I/O until the request has been completely addressed. Most small requests are handled before the threshold is reached, so they are read from and written to cache memory and transferred to or from disk asynchronously, in larger, better aligned units. But larger requests are handled directly once the threshold is exceeded, without extra caching overhead.

Direct I/O

The following samu commands let you change the default I/O behavior of a specified family set device from paged I/O to direct I/O and vice versa:

The commands below control how sparse files are initialized during direct I/O:

Configuring Single-Writer/Multi-Reader File Systems

The commands below control how and when a read-only host obtains metadata updates from the metadata server.

Configuring File Systems for Oracle Real Application Clusters (RAC)

The following commands are specific to implementations of Oracle Real Application Clusters (RAC) that store data on Oracle HSM file systems using asynchronous I/O and Solaris Volume Manager mirrored volumes.

Working with Service and Support Representatives

The samu interface includes a number of commands that are designed to help you assist Oracle technical support and service representatives. These handle the following tasks: