Improving Throughput to Storage Media
Increasing Job Manager Update Interval
Optimizing Collect File Creation
Increasing Data Transfer Throughput From Clients
Configuring Windows Servers for Better Performance
Configuring for Best Performance of Background Services
Optimizing Data Movement Performance
Using Multiplexing Options to Improve the Data Movement Performance
Optimizing Data Movement Performance for the Encrypted Data
Optimizing Performance of Shared Index Cache
Choosing the location of Index Cache
Creating Shared Index Cache for Alternate Data Paths (GridStor)
Cloud Connection Performance Tuning
Improving Speed of Restore Operations
Performance Tuning for 10GB Ethernet Environments
You can improve the performance of data movement operations using the parameters described below. Use one or more parameters to improve the performance.
The throughput to a storage media depends on the speed at which chunks are written to that media and the number of entries per chunk in the Index Cache. Increasing the CHUNK SIZE improves the throughput to a storage media.
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.
By default, the chunk size is configured to get the optimal throughput to the storage media.
The following table gives default chunk size for tape and disk backups and the recommended range of chunk size:
Storage Media | Backup Type | Default Chunk Size | Recommended Range of Chunk Size |
Tape | Granular Backup | 4 GB | 8 GB, 16 GB, 32 GB |
Database Backup | 16 GB | 8 GB, 16 GB, 32 GB | |
Disk and Optical | All Backups | 2 GB | ---- |
Direct Attached NDMP | All backups | 4 GB | ---- |
Do not reduce the default chunk size if you want to perform data movement using NDMP Remote Server (NRS). |
For more information about modifying the chunk size, refer to Increasing Chunk Size.
Use the following registry key to modify the chunk size:
Registry Key | Location | Value |
DMMBchunksize |
|
Chunk size in MB |
Increase the block size for faster write operations to tape media. The default block size for write operation is 64 KB.
For more information about modifying the block size, refer to Increasing Block Size. You can also use setFlushBlockSize tool available in the Resource Pack to modify the block size.
Before increasing the block size, ensure that the following criteria are satisfied:
If you want to modify the block size, use the following supported values of block size:
MediaAgent | Supported Block Size |
Windows 2000 MediaAgent with direct attached SCSI devices | 64 KB or lower |
ONTAP 6.5 NetApp Appliances | 64 KB or lower |
DART 5.3 EMC filers | 128 KB or lower |
The Collect File records the path and name of each scanned file that is included in the backup. When you are performing backup of a large data, a large collect files are generated. You can divide the content in more number of collect files and reduce the time taken by a data reader to read the collect file.
By default, the number of collect files is equal to twice the number of data readers (2 x Number of Data Readers). 2 is a multiplication factor which is the number of Collect files that will be created for the job. You can change the multiplication factor.
Use the following registry key to modify the multiplication factor:
Registry Key | Location | Value |
For_Multiple_Reads_On_Disk_Collect_Split_Multiplication_Factor | Create the Registry Key in a Windows client at the following location: HKEY_LOCAL_MACHINE\SOFTWARE\CommVault Systems\Galaxy\Instance<xxx>\ |
2 or more |
The data transfer from the client to media agent can be increased using the following methods:
The Network Agents establish data pipes to transfer data from the client to the MediaAgent. Therefore, increasing the number of Network Agents will increase the data transfer throughput from the client.
If required, modify the other network parameters of the client, such as network card speed, network switch speed etc., to get the optimum data transfer throughput. |
Follow the steps given below to increase the data transfer throughput:
If you increase the number of network agents on a MediaAgent, more CPU cycles and RAM are consumed. This will result into less number of backups getting completed on the MediaAgent. If you are expecting additional backup operation on a MediaAgent, it is recommended to set Network Agents to 1.
The Data pipe buffers determine the amount of shared memory allocated on each computer for data pipes. The size of each buffer is 64K. By default, 30 data pipe buffers are established on each server for data movement operations. You can increase the data transfer throughput from the client by increasing the number of data pipe buffers.
Use the following registry key to modify the data pipe buffers:
Registry Key | Location | Value |
nNumPipelineBuffers |
|
60, 120, 150, 300 |
When you increase the number of data pipe buffers, more shared memory is consumed by the client or MediaAgent. This may degrade the server performance. Therefore, before increasing the number of data pipe buffers, ensure that the adequate shared memory is available. You can optimize the number of data pipe buffers by monitoring the number of concurrent backups completed on the server.
When the MediaAgent and CommServe are installed on Windows Server 2003 or 2008, follow the steps given below to improve the performance of background services:
|
The following sections provide guidelines to improve the performance of backup and restore operations.
Follow the steps given below to improve the data movement for a File System iDataAgent:
If the Storage Policy is not configured for multiple data streams, then subclients will perform the data protection operations one after the other in a serial.
For the HP-UX 11i v3 file system, do the following steps to improve the performance of backup operations.
The select(2) system call enhancement is delivered through the PHKL_41700 patch.
kctune hires_timeout_enable=1
The select(2)code has been modified to use the high resolution timer functionality. The applications which use select(2)and specify small timeouts will timeout right after the specified duration and this will reduce the long wait time. With this higher resolution, the timer API interfaces can support time intervals less than the current 10 ms. The hires_timeout_enable parameter changes take effect immediately. So it can be done while the server is up and running.
You can improve the performance of backup and restore operations by modifying the following parameters for SQL iDataAgent:
Parameter | Default Value | Supported Values |
maximum size of data blocks used during backups | 64KB | 2 to 512 KB. The value must be power of 2 |
number of buffers used to reserve bandwidth for data transfer | 1 | 1,2,3 |
maximum amount of data to be transferred at a time during backups | 64KB | 64KB to 4 GB The value must be a multiple of 64 |
For more information on how to setup these parameters, refer to Enhancing Performance During Backups.
Exchange Mailbox data movement operations are constrained by the Microsoft MAPI API which limits data movement operations to an average of 2GB per hour per stream. This mode of data movement actually logs into each mailbox and scans each piece of mail as it is delivered to the Inbox.
You can improve the performance of data movement operations by creating multiple subclients. This will enable parallel data movement operations. However, the cumulative throughput of the Exchange server may decrease if sufficient system resources are not available. Therefore create user-defined subclients to divide mailbox backups into groups based on specific criteria. For more information, refer to Grouping Mailboxes Into User-Defined Subclients.
You can optimize the performance of data movement operations using the following methods:
If the disk is highly fragmented disk the size of the data in backup is significantly larger than the valid data on the client computer. This will increase the time required for data movement operation and also consume more space on the storage media.
The Image Level iDataAgent backs up extents on the source drive. The default value is 512KB. You can change the disk extent size to improve the performance of data movement operations. The supported values are 64KB, 128KB, 256KB, 512KB and 1 MB. The value must be power of 2.
Consider the size of the volume to be protected by the Image Level iDataAgent relative to the size of the extent to be used. For example, a 512KB extent setting against a backup of 10KB of data is backed up entirely, including all of the empty blocks of data (502KB of data).
For more information, refer to Change the extent size for Backup Applications.
When you are changing the extent size, keep in mind that:
The Image Level iDataAgent backs up extents on the source drive. The default value is 2MB (4096 X 512 bytes). You can change the disk extent size to improve the performance of data movement operations. The supported values are 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768, etc. The value must be power of 2.
Consider the size of the volume to be protected by the Image Level iDataAgent relative to the size of the extent to be used. For example, a 512KB extent setting against a backup of 10KB of data is backed up entirely, including all of the empty blocks of data (502KB of data).
You can change the Disk Extent Size while configuring the CXBF device. For more information, refer to Configure a CXBF Device in Volume Explorer.
You can also use the following registry key to change the disk extent size:
Registry Key | Location | Supported Values |
DsnapChunkSize | /etc/CommVaultRegistry/Galaxy/Instance<xxx>/UnixImageIDA/.properties | 128, 256, 512, 1024, 2048, 4096, 8192, 16384, 32768, etc. The value must be power of 2. |
When you are changing the extent size, keep in mind that:
You can improve the performance of backup and restore operations by modifying the following parameters for Oracle iDataAgent:
Parameter | Default Value | Description |
Number of data files and archive files to be included in each RMAN backup set | 32 | Increase the value of this parameter for faster backups. However, increasing the value in this parameter will create additional CPU overhead in order to optimize the population of the Archive file. The granular restore of individual table spaces and data files will be slower. |
Number of open data files that can be read by RMAN during the backup | 8 | Reduce the value of this parameter for faster backup and less system load on the Oracle server. |
Maximum size of data blocks used during backups | 262144 KB | Increase the value of this parameter for faster tape drives or magnetic
libraries. For Example: For SDLT or LTO tape drives, a minimum value of 1048576 (1MB) is recommended. |
For more information on how to setup these parameters, refer to Enhancing Backup Performance.
Also, if you are experiencing performance issues during backup, you can troubleshoot them by enabling logging of performance details in the log files. These performance counters contain information that help in resolving the performance related issues during backups. See Troubleshooting Performance Issues for more information.
You can optimize the performance by modifying the following parameters for Informix iDataAgent:
You can modify the buffer size settings using the BAR_XFER_BUF_SIZE parameter in the $ONCONFIF file on the database server. For best performance of data movement operations set the value to 30.
You can also configure the following parameters on the Informix database server:
Parameter | Description |
BAR_IDLE_TIMEOUT | This parameter specifies the maximum number of minutes that an ON-Bar worker process is idle before it is shut down. |
BAR_MAX_BACKUP | This parameter specifies the maximum number of backup processes per ON–Bar command. You can control parallel execution of ON–Bar processes using this configuration parameter. |
BAR_NB_XPORT_COUNT | This parameter specifies the number of shared-memory data buffers for each backup or restore process. |
BAR_PROGRESS_FREQ | This parameter specifies how frequently the backup or restore progress messages display in the activity log. |
LTAPEBLK | This parameter specifies the block size for logical-log backups made with ontape, onload, onunload. |
LTAPEDEV | This parameter specifies the tape device used for logical-log backups made with ontape. |
TAPESIZE | This parameter specifies tape size for logical-log backups made with ontape. |
You can improve the data movement performance by reviewing the following parameters in the Netware Server environment:
Verify the Network Interface Card (NIC) settings for the Novell Client severs, MediaAgents, CommServe, Routers, Network Switches. Client server side NIC settings are found in MONITOR.NLM, although the name of the statistic varies between LAN drivers.
Ensure that each port is configured for Full Duplex connectivity. Half-Duplex network connections create an increase in IP stack collisions and reduces data movement efficiency.
You can improve the performance of backup operation by factor of 2x or 3x by using the Novell's Storage Management Services (SMS). For more information, refer to Optimize Backups Using Novell's Storage Management Services (SMS).
You can tune the Novell Netware v6.5 NSS volumes for better data movement operations by adjusting the parameters mentioned below. It is recommended to consult with the Netware Administrator before adjusting any parameters for Novell Netware v6.5 NSS volume. For more information about Best Practices for NSS volume performance tuning, refer to http://support.novell.com/cgi-bin/search/searchtid.cgi?/10065215.htm.
You can use the following file to update these parameters:
C:\NWSERVER\NSSSTART.CFG
Parameter | Recommended Value | Description |
NameCacheSize | 2000 | |
OpenFileHashShift | 15 | You can increase the value of this parameter up to 17 if 1GB of memory is available. |
CacheBalance | 80 | The Cache Balance refers to Cache Balance Percent. You can find the
Cache Balance Percent in Monitor | Server Parameters |
Novell Storage Services. The Cache Balance configured on the NetWare client server should be equal to the percent of total disk space mounted on the server that belongs to NSS. |
ClosedFileCacheSize | 100000 | You can increase the value of this parameter up to 100000 if 1GB of memory is available. |
Use the /nosalvage switch on GroupWise volumes to prevent the volumes from filling with deleted message files. This will give similar effect as PurgeImmediate attribute on traditional volumes. The FileFlushTimer and BufferFlushTimer switches should not be used for data movement operations. These settings can stagnate migration of clustered volumes and may result into data loss in the event of an abnormal termination on the server.
You can improve the performance of data movement operation for NDMP and NRS clients using the following guidelines:
netstat -a
The send_q information will determine whether network queues are becoming full.
Follow the guidelines given below to improve the performance of the NetApp Magnetic or Disk Library:
Follow the guideline given below to improve performance of restore operations for NetApp File Server:
You can improve the performance of Auxiliary Copy operation using the following guidelines:
The Auxiliary copy operation enables you to create a secondary copies of data associated with a storage policy. When the data to be copied exceeds a specific threshold or if the data is not copied for specific number of days, you can get a notification. Once you get the notification, you can optimize and troubleshoot the auxiliary copy operation.
You can enable an alert to get the notification and set the thresholds. For more information, refer to Best Practices.
You can reduce the time required for data movement to tape using multiplexing to stream the data. You can determine the multiplexing factor by analyzing the network configuration and the required disk throughput.
Typical multiplexing factors are set between 2 and 5 on 100BaseT networks.
Typical multiplexing factors are set between 5 and 8 on 1000BaseT networks.
For more information, refer to Data Multiplexing.
The encrypted data movement operation is always slower than an unencrypted data movement operation. The software based encryption consumes CPU resources. Therefore, it is recommended to perform the encryption of the data on the system that has the maximum CPU resources available.
You can select the method of encryption from several algorithms and key length options. For more information about Configuration options, refer to Data Encryption.
It is recommended to select specific data from a client for encryption, instead of performing bulk encryption for the client. Do not encrypt Windows executable and programs and similar object types.
Use the following guidelines to improve the performance of shared index cache:
The size of shared index cache depends on the condition of the file system on which the Shared Index Cache resides and also the input/output supported by the MediaAgent. For large cache, the Disk I/O the becomes an issue. Therefore, RAID0, RAID1, or RAID5 are preferred schemes to define a speedy location for the cache. RAID 1 is the fastest and safest. RAID 5 is fast and inexpensive.
Follow the guidelines listed below while creating a Shared Index Cache Directory for alternate data paths:
If the cloud connectivity is via very high speed network of about 1 GB per second or higher, follow the below recommendations to get the maximum performance for the cloud based backup/restore/auxcopy jobs.
These settings will consume large amounts of memory on the MediaAgents and and an x64 MediaAgent is mandatory to support it. Hence these settings are not recommended for low speed networks. |
Following are the Cloud performance specific Registry Keys for cloud based backup jobs:
Use the below steps to enable/disable the temp file for upload. Value 0 is for memory buffer.
|
Use the below steps to set the maximum number of concurrent upload thread per stream.
|
Use the below steps to set the size of the subfile to upload to cloud server.
|
|
Following are the Cloud performance specific Registry Keys for cloud based restore jobs:
Use the below steps to enable/disable the temp file for upload. Value 0 is for memory buffer.
|
|
|
|
You can select the following overwrite options from the Restore Options dialog box.
Whenever you select any of these options, large files on the destination computer will be deleted. If set the following option ON, the operating system on the destination computer will perform bulk deletion and not the sequential deletion:
SET IMMEDIATE PURGE OF DELETED FILE=ON
Follow the guidelines given below to improve the performance of data movement in a 10GB Ethernet environment: