locked
Load Balance Distribution Points RRS feed

  • Question

  • I think there is an error or at least a misleading statement in the documentation for Distribution Points.

    The documentation says that the DPs with the package are sorted according to an algorythm.  This includes how near the DP is to the client in a network map. Whether the DP is in the same AD site as the client or not, and whether the DP has BITS enabled or not.  The documentation calls this ranking a "sorting category."  It allows for multiple DPs to have the same sorting category.  The documentation states "Within each sorting category, the order of distribution points is nondeterministic, which provides a level of load-balancing for the servers."  This is incorrect.  When multiple DPs have the same "sorting category" in relation to a group of clients, all clients will access the package from the same DP and no load balancing occurs. The DP list appears to be provided in alphabetical order withing the same sorting rank, and the client chooses the first DP listed alphabetically - not randomly or non-deterministically. 

     

    For example.  Take 3 identical clients on a single subnet and three identically configured DPs on a second subnet (all BITS enabled, same AD site, etc.).  When the client requests the list of DPs for a package, AD will return the same list in the same order to all three clients.  The clients will choose the DP that is first on the list.  All three clients will get the package from the same DP - the first one listed.  The documentation suggests that when all DPs have the same ranking, that the order of the DPs sent to the client is non-deterministic and that the clients will load balance across same ranked DPs.  This is not the case.  All the load will be on the first listed DP.  All clients will get the package from the same DP.  The only reason a client would go to DP #2 is if DP#1 is not available.  There is no level of load balancing provided as stated in the documentation unless it is refering to something akin to failover clustering and not load balancing.  I view load balancing to mean that the load is spread out somewhat equally among several DP resources.  This is not how DP selection works.

    What the documentation says at http://technet.microsoft.com/en-us/library/bb632366.aspx

    Within each sorting category, the order of distribution points is nondeterministic, which provides a level of load-balancing for the servers. So, a client would attempt to connect to any of the distribution points that are enabled with the option Allow clients to transfer content from this distribution point using BITS, HTTP, and HTTPS on the same IP subnet as the client, and then attempt to connect to any of the distribution points that are not enabled with this option when they are on the same IP subnet as the client.

    Internet-Based Distribution Points

    When the client is being managed over the Internet, and requesting content from a site configured for Internet-based client management, all Internet-based distribution points in that site will be returned to the client if they contain the content requested by the client.

    In this scenario, distribution points must be configured with the option Allow clients to transfer content from this distribution point using BITS, HTTP, and HTTPS, and the selection of a distribution point by the client is nondeterministic.

     

    • Moved by TorstenMMVP Monday, July 18, 2011 5:13 PM moved to the Documentation subforum (From:Configuration Manager Software Distribution)
    Monday, July 18, 2011 5:07 PM

Answers

  • Send this to smsdocs@microsoft.com

     

     


    John Marcum | http://myitforum.com/cs2/blogs/jmarcum/|
    Monday, July 18, 2011 5:22 PM
  • Also, I have reported it to our feedback team.

    Thanks. :)


    Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.
    Friday, August 5, 2011 10:05 AM

All replies

  • Send this to smsdocs@microsoft.com

     

     


    John Marcum | http://myitforum.com/cs2/blogs/jmarcum/|
    Monday, July 18, 2011 5:22 PM
  • Also, I have reported it to our feedback team.

    Thanks. :)


    Please remember to click “Mark as Answer” on the post that helps you, and to click “Unmark as Answer” if a marked post does not actually answer your question. This can be beneficial to other community members reading the thread.
    Friday, August 5, 2011 10:05 AM
  • Hi All,

    I would like to know the status of the above conflicts..

    is Todd Miller statement point proved or not?

    Tuesday, January 8, 2013 3:49 PM
  • I would have sworn I came back to update this thread - sorry.

    My observations were correct, but I drew the wrong conclusions.  The problem turned out to be that the IP subnet that the distribution points were on was not listed in my site boundaries.    While the client subnets were in the boundaries, the site servers, sitting in our Data Center, were not within the site boundaries.

    Once I added the IP addresses of the distribution points so that they were also within the site boundaries, the pseudo-round robin mechanism worked.  The list of DPs started to be returned in a random order.

    When the DP's IP addresses were not themselves in the site boundaries, the DP lists for package sources were returned in alphabetical order and not randomized.

    • Proposed as answer by Garth JonesMVP Tuesday, January 8, 2013 6:12 PM
    Tuesday, January 8, 2013 5:01 PM