ITPub博客

首页 > Linux操作系统 > Linux操作系统 > Performance Tunables for Media Management

Performance Tunables for Media Management

原创 Linux操作系统 作者:lhl1212 时间:2011-06-09 10:13:48 0 删除 编辑

Media Management > Operations > Performance Tunables for Media Management

Published 11/19/2010 02:50 PM   |    Updated 11/19/2010 02:50 PM

Performance Tunables for Media Management

Topics | How To | Related Topics


Increasing Chunk Size

Increasing Block Size

Increasing Job Manager Update Intervals

Unbuffered IO For Magnetic Libraries

Data Multiplexing


The following sections describe some of the tunable parameters in Media Management that can be used to improve the performance.

Increasing Chunk Size

A chunk is the unit of data that the MediaAgent software uses to store data on media. For Sequential access media, chunk is defined as data between two file markers. The default chunk size for File System like data is 4 GB and for data associated with databases is 16 GB. For Random access media each chunk is a file on the disk. The default chunk size for this type of media is 2 GB.

This parameter will have an impact only when backing up to tape. A higher chunk size will give you better throughput. Recommended values are: 8 GB, 16 GB or 32 GB. However the disadvantage is that granular restores (e.g., single file restore) will be slower. On the other hand large restores, like a full machine rebuild will be a bit faster.

 A lower value is recommended for frequent checks of slower data protection operations, especially when data is moving across a WAN link.

Chunk size is configurable using the Chunk Size tab of the Media Management Configuration dialog box, on a per application basis. (See Set the Chunk Size for Each Agent for step-by-step instructions.)

The chunk size established in this dialog box affects all data write operations in the CommCell. However you can establish the DMMBCHUNKSIZE registry key to control the chunk size of data write operations going to the MediaAgent in which the registry key is created. Note that if this registry key is created it will override the values established in this dialog box for that MediaAgent.

Also note that the values established for the NAS NDMP iDataAgents in this dialog box only apply to libraries attached to the MediaAgent using NDMP Remote Server. For libraries attached to a File Server, the system, by default, uses 4GB as the chunk size for file servers that support restartability (e.g., NetApp filer). For such file servers you can establish the nRESTARTWINDOWSIZE registry key to control the chunk size. For other file servers, the chunk size is based on the size of the tape used to backup the data and this is not configurable.

(The values in the Chunk Size tab of the Media Management Configuration dialog box applies only to tape media; for magnetic and optical media the system by default uses 2 GB as the chunk size.)

Chunk Size can also be established from the Data Path Properties dialog box available from the Data Paths tab of the Copy Properties dialog box, for the specific data path. (See Set the Chunk Size and Block Size for a Data Path for step-by-step instructions.)

Chunk size established at the data path level overrides the chunk sizes established for the various Agents, as discussed in the preceding paragraph.


Increasing Block Size

MediaAgents can write to media using different block sizes, if the Operating System associated with the MediaAgent in which the library is configured supports a higher block size. The system can write block sizes up to 256 KB and can automatically read block sizes up to 512 KB. If the block sizes are larger than 512 KB, read operations from the media will fail. Also note that such media will be over written and re-used if the When Content Verification Failed option is enabled in the Library Properties (Media) dialog box

Block sizes can be modified from the Data Path Properties dialog box available from the Data Paths tab of the Copy Properties dialog box, for the specific data path. (See Set the Chunk Size and Block Size for a Data Path for step-by-step instructions.)

Higher block sizes on tape media are faster, but consider the following before increasing the block size:

  • Note that the new block size must be supported by the HBA driver as well as by the tape drive. The most commonly used values are 128 KB or 256 KB.
  • MediaAgents with direct attached SCSI devices on Windows 2000 do not support block sizes higher than 64 KB.
  • NetApp NAS file servers for ONTAP 6.5.3 or later allow block sizes up to 256 KB, but NetApp servers for ONTAP 6.5.2 or earlier allow maximum block sizes of 64 KB.
  • EMC Celerra NAS file servers support block sizes up to 128 KB.
  • If you do specify a block size larger than the allowed limit, the NAS backup process automatically changes your entry to the maximum allowable block size.

For magnetic mount paths, block size is the buffer size used for each write operation. (The File system allocation unit size is decided at the time of formatting the volume.)

Cautions:

  • All MediaAgents that share a storage policy must support the higher block size.
  • If the MediaAgent that performs the restore is different than the original MediaAgent, then the destination MediaAgent must have HBA cards and drives to be able to read the higher block size.

Increasing Job Manager Update Intervals

The system, by default updates the Job Manager, every 5 minutes or whenever a chunk is closed, whichever is sooner. You can marginally improve the performance by increasing the job update interval when you increase the chunk size. However increasing the update interval will result in infrequent updates to the job progress in the Job Controller. These changes will not impact NAS NDMP iDataAgent backups.

 (See Job Update Interval for more information on modifying the settings.)


Unbuffered IO For Magnetic Libraries

When backing up to magnetic devices, you can enable Use Unbuffered I/O option in the Mount Path Properties dialog box to bypass the Windows file system buffering. Varying amounts of speedup will be observed depending upon the architecture of the magnetic disk. (See Use Unbuffered I/O for additional information.)

You can establish the dwMaxAsyncIoRequests registry key to control the number of read ahead buffers in the unbuffered I/O mode.


Data Multiplexing

Multiplexing does not improve performance of an individual backup operation. However multiple backups run in parallel to a single tape drive, results in better utilization of the tape drives, especially when the backups are from slower clients. This helps in better overall throughput and reduction in the backup window.

For LAN backups, make sure that the network between the clients and MediaAgent is capable of supporting multiple simultaneous backups.

Do not over multiplex. That would be counter-productive and slow down the backups as well as restores. Multiplexing factor must be set equal to the ratio of tape drive throughput and client source speed. For example, if the tape drive has rated speed of 40 Mb/sec and clients are able to supply the data at about 12 Mb/sec, then a multiplexing factor of 3 is advisable. Typical multiplexing factor is between 2 and 5.

Also note that restores from multiplexed data are not slower.

See Data Multiplexing for complete details.

来自 “ ITPUB博客 ” ,链接:http://blog.itpub.net/611609/viewspace-697453/,如需转载,请注明出处,否则将追究法律责任。

请登录后发表评论 登录
全部评论

注册时间:2008-12-08

  • 博文量
    252
  • 访问量
    496559