14 Responses

  1. Renato
    Renato at |

    As per the latest (Jan 2015 EMC VNX TechBook version 4.0 Rev.5)

    h8229-vnx-vmware-tb.pdf

    the multiple subnet topology is recommended. I have personally opened cases and being told by support to explain why is there a single ISCSi subnet on some specific clients…and if there is a possibility to turn it into multiple subnets.

    i can publish the pdf in dropbox and share the link if you would like

    Reply
  2. [VMware] news + interesting reads from the blogosphere: 09 Mar 2015

    […] When To Use Multiple Subnet iSCSI Network Design http://wahlnetwork.com/2015/03/09/when-to-use-multiple-subnet-iscsi-network-design/ […]

  3. When To Use Multiple Subnet iSCSI Network Design | Storage CH Blog

    […] Read on here […]

  4. lukedudney
    lukedudney at |

    NetApp are still recommending the multiple-subnet design as of May 2014 :

    From http://www.netapp.com/us/media/tr-4068.pdf pp113
    The Data ONTAP operating system can use multiple sessions across multiple networks. When designing the iSCSI storage network, NetApp recommends using multiple VMkernel network interfaces on different network subnets…

    By default, ALUA is enabled for all iSCSI LUNs serviced by clustered Data ONTAP. NetApp recommends using portsets to mask extraneous paths to the LUNs and prevent path exhaustion for the VMware hosts.

    Using the Virtual Storage Console to manage VMware storage automatically applies the NetApp best practices for provisioning and presenting LUNs to initiators

    Reply
  5. jhope188@gmail.com
    [email protected] at |

    So Chris, Appreciate all your help here. I will say iSCSI Best practice is still something that plagues me. (Also thanks for the great CCNA-DC training on Pluralsight!!) Anywho…

    I know this is a netapp design and maybe its due to the differences with EMC, but even going off VMware best practices I thought 2 Subnets was preferred. In addition to that the whole single vSwitch to VMK part gets a little confusing. Recently we have just been doing single vswitch to vmk for the iscsi binding based on a vlan. So we have single vswitch with vmk on vlan100 going to san and then vswitch with second vmk on vlan200 going to san. The difference in our deplyment and what is represented above is the whole active passive situation.

    The confusion here is the meaning behind the word. When I think Active/Active im thinking VMAX or XIO not VNX or VNXe. Then EMC one day was like no your going about it all wrong. A VNXe while not an active active array, can have luns addressed by both SP’s just only one SP can own the lun at a time. So when we deploy luns we create luns owned by both SPA and SPB during deployment. Technically they can be accessed simultaneously just only one of the SP’s can own that lun.

    At that point the last diagram you had would look more like our setups except IO would be serviced simultaneously from both VMK interaces on both SP’s. Lun 101 and 102 would live on SPA and lun 103 and 104 would live on B.

    Again appreciate all you awesome training and definitely a great article on the topic.

    Reply
  6. Jon Hope
    Jon Hope at |

    cool beans. So at the end of the day were on the same page. I think I need a refresher on the VMWare networking best practices. Gonna go buy the Networking for VMWare administrators book.

    Appreciate all the great training you provide. Owe the first part of my CCNA-DC to your training on Pluralsight.

    Reply
  7. Boudewijn Plomp
    Boudewijn Plomp at |

    I always prefer to use separate subnets (VLANs). Because of the following reasons:

    NetApp:
    First of all NetApp requires multiple paths. If you ignore this and try to use MPIO with a single subnet, you get the following warning and error on your NetApp:
    – iSCSI Notice: New session from initiator iqn.2013-01.com.ucs:company-srv01 at IP addr 10.20.30.1
    – iSCSI Warning: ISCSI: New session request from initiator iqn.2013-01.com.company-srv01, a session from this initiator already exists.
    And iSCSI performance does not seem to work optimal. So like Chris mentions, check you vendor documentation.

    Windows Failover Cluster Manager:
    If you run the Cluster Validation Wizard you receive a warning it will warn you two network interfaces (iSCSI) are using the same subnet. So your cluster is not fully validated.

    Windows MPIO:
    When you configure an iSCSI target the initiator source IP Address is by default is 0.0.0.0. If you have two iSCSI interfaces, you don’t know which interfaces is used for outbound connection. Investigation on switch traffic revealed me that optimal paths are not used properly. So you have to configure multiple connections you have to configure each source IP Address because you are in the same subnet. One more reason for my why separate subnets (VLANs) are much more preferred.

    Switch Uplink Connectivity:
    Switches sometimes have uplinks to other switches, often with cross connectivity. This can cause a non-optimal path. But if you separate the subnets you can control which path to use.

    Reply

Share your point of view!