Topic
  • 10 replies
  • Latest Post - ‏2011-02-21T20:32:10Z by m_strnad
m_strnad
m_strnad
7 Posts

Pinned topic mmbackup and multiple TSM servers

‏2011-02-02T14:19:32Z |
Hi all,

I have some questions related to usage of multiple TSM servers with mmbackup.

The docu Says

"The mmbackup command supports backing up of GPFS file system data to multiple Tivoli Storage Manager servers. The ability to partition file backups across multiple TSM servers is particularly useful for installations that have a large number of files. For information on setting up multiple TSM servers, see Tivoli Storage Manager requirements."

and in the Tivoli Storage Manager requirements section only:

"If using multiple TSM servers to back up a GPFS™ file system, ensure that the TSM servers are set up properly."

But what exactly does this mean?

I suppose:
1) TSM client installed and configured on every node running mmbackup.
2) Each node using the same TSM nodename but connecting to different TSM servers.

But then:
1) Does GPFS/mmbackup keep trace of which file was sent to which TSM server / by which GPFS node?
2) If so, what happens when a node eventually needs to be replaced because of HW failure?
3) If not, how is the expiration of a file achieved if a different GPFS node (connected to different TSM server) should handle file previously backed up by some other GPFS node.

Thank you in advance for any useful information.

Michal Strnad
Updated on 2011-02-21T20:32:10Z at 2011-02-21T20:32:10Z by m_strnad
  • sberman
    sberman
    2 Posts

    Re: mmbackup and multiple TSM servers

    ‏2011-02-02T16:03:16Z  
    Michal,
    In general "set up properly" will mean that each TSM server will be designated to back up some part or all of a GPFS file system. In the case of "some part" of a GPFS, the TSM server will need to partition the file system according to some criteria. For example: GPFS file system called "gpfs1" and mounted at "/gpfs1" has large directories in the top level called
    /gpfs1/projects
    /gpfs1/users
    /gpfs1/archives
    Further, TSM servers called tsmarch1 and tsmarch2. In the dsm.sys file on every node that has TSM BA Client installed, there will be "server stanzas" set up for each of these TSM servers. Further include and exclude rules can be defined for each server. The way to partition the work-load will be to record in dsm.sys a set of include/exclude rules for each server so that it excludes some or all of the file system contents that you do not want on that server. Everything else would naturally be backed up on that server. Using "exclude" will permit other top-level directories to be added later and ensured they get backed up, and so is preferable to "include" rules.

    Example dsm.sys:
    ***
    • tsmarch1 backs up "archives"
    ***
    SErvername tsmarch1
    COMMMethod TCPip
    TCPPort 1500
    TCPServeraddress 192.168.1.171
    passwordaccess generate
    NODEname node1
    PASSWORD password
    exclude.dir /gpfs1/projects
    exclude.dir /gpfs1/users
    ***
    • tsmarch2 backs up "projects" and "users"
    ***
    SErvername tsmarch2
    COMMMethod TCPip
    TCPPort 1500
    TCPServeraddress 192.168.1.172
    passwordaccess generate
    NODEname node1
    PASSWORD password
    exclude.dir /gpfs1/archives
    During mmbackup execution, on any node with such dsm.sys configuration, the mmbackup scan of the file systems will utilize the "exclude.dir" rules in the dsm.sys for the server it is using and obey that directive. Everything not excluded will be a candidate for backup to that server. Mmbackup loops through all servers specified on the command line with the --tsm-servers TSMserverhttp://,TSMserver... option.

    Hope this helps!
    -S
  • m_strnad
    m_strnad
    7 Posts

    Re: mmbackup and multiple TSM servers

    ‏2011-02-02T20:29:10Z  
    • sberman
    • ‏2011-02-02T16:03:16Z
    Michal,
    In general "set up properly" will mean that each TSM server will be designated to back up some part or all of a GPFS file system. In the case of "some part" of a GPFS, the TSM server will need to partition the file system according to some criteria. For example: GPFS file system called "gpfs1" and mounted at "/gpfs1" has large directories in the top level called
    /gpfs1/projects
    /gpfs1/users
    /gpfs1/archives
    Further, TSM servers called tsmarch1 and tsmarch2. In the dsm.sys file on every node that has TSM BA Client installed, there will be "server stanzas" set up for each of these TSM servers. Further include and exclude rules can be defined for each server. The way to partition the work-load will be to record in dsm.sys a set of include/exclude rules for each server so that it excludes some or all of the file system contents that you do not want on that server. Everything else would naturally be backed up on that server. Using "exclude" will permit other top-level directories to be added later and ensured they get backed up, and so is preferable to "include" rules.

    Example dsm.sys:
    ***
    • tsmarch1 backs up "archives"
    ***
    SErvername tsmarch1
    COMMMethod TCPip
    TCPPort 1500
    TCPServeraddress 192.168.1.171
    passwordaccess generate
    NODEname node1
    PASSWORD password
    exclude.dir /gpfs1/projects
    exclude.dir /gpfs1/users
    ***
    • tsmarch2 backs up "projects" and "users"
    ***
    SErvername tsmarch2
    COMMMethod TCPip
    TCPPort 1500
    TCPServeraddress 192.168.1.172
    passwordaccess generate
    NODEname node1
    PASSWORD password
    exclude.dir /gpfs1/archives
    During mmbackup execution, on any node with such dsm.sys configuration, the mmbackup scan of the file systems will utilize the "exclude.dir" rules in the dsm.sys for the server it is using and obey that directive. Everything not excluded will be a candidate for backup to that server. Mmbackup loops through all servers specified on the command line with the --tsm-servers TSMserverhttp://,TSMserver... option.

    Hope this helps!
    -S
    Thanks sberman. Your explanation is just clear enough for me.
    I was not aware of the way mmbackup handles dsm.sys content.
  • gcorneau
    gcorneau
    4 Posts

    Re: mmbackup and multiple TSM servers

    ‏2011-02-02T21:10:52Z  
    • m_strnad
    • ‏2011-02-02T20:29:10Z
    Thanks sberman. Your explanation is just clear enough for me.
    I was not aware of the way mmbackup handles dsm.sys content.
    A slight alternative to what Scott suggested (this is computing: there's always more than one way to do things, and all of them can be right!) is to utilize the asnodename configuration option in TSM and let the mmbackup command backup the entire "/gpfs1" file system, divvying up the sections/files to backup based on the number of clients specified in the mmbackup configuration file:

    http://www.ibm.com/developerworks/wikis/display/hpccentral/GPFS+and+TSM+Sample+Scenario
    http://publib.boulder.ibm.com/infocenter/tivihelp/v1r1/topic/com.ibm.itsmfdt.doc/ans50000128.htm#gpfs


    Glen Corneau
    IBM Power Systems Advanced Technical Skills
  • m_strnad
    m_strnad
    7 Posts

    Re: mmbackup and multiple TSM servers

    ‏2011-02-20T17:41:07Z  
    • sberman
    • ‏2011-02-02T16:03:16Z
    Michal,
    In general "set up properly" will mean that each TSM server will be designated to back up some part or all of a GPFS file system. In the case of "some part" of a GPFS, the TSM server will need to partition the file system according to some criteria. For example: GPFS file system called "gpfs1" and mounted at "/gpfs1" has large directories in the top level called
    /gpfs1/projects
    /gpfs1/users
    /gpfs1/archives
    Further, TSM servers called tsmarch1 and tsmarch2. In the dsm.sys file on every node that has TSM BA Client installed, there will be "server stanzas" set up for each of these TSM servers. Further include and exclude rules can be defined for each server. The way to partition the work-load will be to record in dsm.sys a set of include/exclude rules for each server so that it excludes some or all of the file system contents that you do not want on that server. Everything else would naturally be backed up on that server. Using "exclude" will permit other top-level directories to be added later and ensured they get backed up, and so is preferable to "include" rules.

    Example dsm.sys:
    ***
    • tsmarch1 backs up "archives"
    ***
    SErvername tsmarch1
    COMMMethod TCPip
    TCPPort 1500
    TCPServeraddress 192.168.1.171
    passwordaccess generate
    NODEname node1
    PASSWORD password
    exclude.dir /gpfs1/projects
    exclude.dir /gpfs1/users
    ***
    • tsmarch2 backs up "projects" and "users"
    ***
    SErvername tsmarch2
    COMMMethod TCPip
    TCPPort 1500
    TCPServeraddress 192.168.1.172
    passwordaccess generate
    NODEname node1
    PASSWORD password
    exclude.dir /gpfs1/archives
    During mmbackup execution, on any node with such dsm.sys configuration, the mmbackup scan of the file systems will utilize the "exclude.dir" rules in the dsm.sys for the server it is using and obey that directive. Everything not excluded will be a candidate for backup to that server. Mmbackup loops through all servers specified on the command line with the --tsm-servers TSMserverhttp://,TSMserver... option.

    Hope this helps!
    -S
    I have one more simple question related to this:

    Does mmbackup obey also the domain options in the dsm.sys config file while deciding which TSM node to use to back up a file?
    Or only include/exclude statements are recognized?
  • m_strnad
    m_strnad
    7 Posts

    Re: mmbackup and multiple TSM servers

    ‏2011-02-20T17:41:57Z  
    Another option came up which I would like to clarify
  • gcorneau
    gcorneau
    4 Posts

    Re: mmbackup and multiple TSM servers

    ‏2011-02-21T14:05:07Z  
    • m_strnad
    • ‏2011-02-20T17:41:07Z
    I have one more simple question related to this:

    Does mmbackup obey also the domain options in the dsm.sys config file while deciding which TSM node to use to back up a file?
    Or only include/exclude statements are recognized?
    Details on mmbackup and include/exclude files can be found in the GPFS Administration Guide:

    http://publib.boulder.ibm.com/infocenter/clresctr/vxrx/topic/com.ibm.cluster.gpfs34.admin.doc/bl1adm_fsmbackup.html


    Glen Corneau
    IBM Power Systems Advanced Technical Skills
  • m_strnad
    m_strnad
    7 Posts

    Re: mmbackup and multiple TSM servers

    ‏2011-02-21T14:28:39Z  
    • gcorneau
    • ‏2011-02-21T14:05:07Z
    Details on mmbackup and include/exclude files can be found in the GPFS Administration Guide:

    http://publib.boulder.ibm.com/infocenter/clresctr/vxrx/topic/com.ibm.cluster.gpfs34.admin.doc/bl1adm_fsmbackup.html


    Glen Corneau
    IBM Power Systems Advanced Technical Skills
    The include/exclude options themselves are clear. Because of the number of subdirectories we do have and TSM nodes we would like to use ti can get quite complex to use them. Therefore we would rather like to use the domain statements together with the virtualmountpoint statements.
    The questions are:

    Does mmbackup during the filelist generation look for the include/exclude statements in dsm.sys? (I guess it does.)
    Is it able to use other methods to "partition" the data to different TSM nodes/servers? (the domain option in particular)

    What is the procedure mmbackup uses to determine which filelist a file belongs to?

    Thank you

    Michal
  • m_strnad
    m_strnad
    7 Posts

    Re: mmbackup and multiple TSM servers

    ‏2011-02-21T14:43:13Z  
    • m_strnad
    • ‏2011-02-21T14:28:39Z
    The include/exclude options themselves are clear. Because of the number of subdirectories we do have and TSM nodes we would like to use ti can get quite complex to use them. Therefore we would rather like to use the domain statements together with the virtualmountpoint statements.
    The questions are:

    Does mmbackup during the filelist generation look for the include/exclude statements in dsm.sys? (I guess it does.)
    Is it able to use other methods to "partition" the data to different TSM nodes/servers? (the domain option in particular)

    What is the procedure mmbackup uses to determine which filelist a file belongs to?

    Thank you

    Michal
    Or does the mmbackup call the "dsmc selective" utility with the same filelist for each of the TSMServers secified in the --tsm-servers option?
    That will let the TSM client decide which files to back up and which not (Different TSM nodes will pick different files from the same filelist).

    That would be good a for the TSM client configuration options (any options specified would aply) but will mean more processing during the "dsmc selective" phase of the backup.
  • sberman
    sberman
    2 Posts

    Re: mmbackup and multiple TSM servers

    ‏2011-02-21T19:40:49Z  
    • m_strnad
    • ‏2011-02-21T14:43:13Z
    Or does the mmbackup call the "dsmc selective" utility with the same filelist for each of the TSMServers secified in the --tsm-servers option?
    That will let the TSM client decide which files to back up and which not (Different TSM nodes will pick different files from the same filelist).

    That would be good a for the TSM client configuration options (any options specified would aply) but will mean more processing during the "dsmc selective" phase of the backup.
    Michal,

    The method mmbackup uses to comply with TSM include/exclude rules is that at the outset, mmbackup performs "dsmc query inclexcl" to get a list of these rules for the server (each server). Using the rules output, it constructs a "Policy Rules Files" that is used by mmapplypolicy to enumerate the file system, guided by the include and exclude rules found. The policy engine filters the rule set to essentially find all objects in the file system subject to these include and exclude directives. This list of "all matching" files is next compared against the last backup recorded for that server in an .mmbackupShadow database file in the root of the GPFS. That comparison yields a set of files changed or new to be backed up, and another list of deleted files to be "expired" from the TSM server. Expiration does not necessarily mean immediate deletion from the server but does mark the entry as "expired" and depending on server config settings will eventually result in deleting the file data and recovering the storage space (tape or disk).

    There are some things to be careful of in using "exclude.dir" statements. The final character of an exclude.dir statement should be the "/" character to ensure that partial matches of similar directory names do not result in omitting directories for which backup was desired.

    -Steve
  • m_strnad
    m_strnad
    7 Posts

    Re: mmbackup and multiple TSM servers

    ‏2011-02-21T20:32:10Z  
    • sberman
    • ‏2011-02-21T19:40:49Z
    Michal,

    The method mmbackup uses to comply with TSM include/exclude rules is that at the outset, mmbackup performs "dsmc query inclexcl" to get a list of these rules for the server (each server). Using the rules output, it constructs a "Policy Rules Files" that is used by mmapplypolicy to enumerate the file system, guided by the include and exclude rules found. The policy engine filters the rule set to essentially find all objects in the file system subject to these include and exclude directives. This list of "all matching" files is next compared against the last backup recorded for that server in an .mmbackupShadow database file in the root of the GPFS. That comparison yields a set of files changed or new to be backed up, and another list of deleted files to be "expired" from the TSM server. Expiration does not necessarily mean immediate deletion from the server but does mark the entry as "expired" and depending on server config settings will eventually result in deleting the file data and recovering the storage space (tape or disk).

    There are some things to be careful of in using "exclude.dir" statements. The final character of an exclude.dir statement should be the "/" character to ensure that partial matches of similar directory names do not result in omitting directories for which backup was desired.

    -Steve
    Thanks Steve.

    This clarifies all.

    Michal