1 Reply Latest reply on Nov 17, 2009 11:23 PM by GregDBA

    Oracle Private Network Design

      I have some questions regarding the Oracle Private Network.

      Basically I am going to have three 3 node clusters. Active, Standby, and Upgrade clusters.

      I am trying to figure out the best way to connect these in a private network. We are using Dell M1000 server chassis and we have ethernet passthrough modules in the I/O slots in the back.

      What does Oracle say about using bonded interfaces for this? I have two options: Option 1 would be to connect these passthroughs to two different Cisco Switches, CoreA and CoreB. This solution would allow for "bonding" in that I could trunk the VLAN across the two switches and NICs could be active/passive. Failover is decent.

      Otherwise I was thinking of getting two Cisco 3750 10/100/1000 switches and stacking them. These switches would allow the NICs to do Link Aggregation Control Protocol (LACP/802.3AD) and get me an aggregated 2Gbps of throughput.

      I could do this same type of bonding on the Core switches but the NICs would have to be plugged into the same Cisco 6500 chassis which would not be good in event of chassis failure.

      Is there any preferred solution or recommendation from Oracle on this? How about user experience? As a Network Engineer I'd rather the best solution dictate my network design rather than my network design dictate a not-as-good solution.

      Thanks in advance for any help you provide. This is my first post here.

        • 1. Re: Oracle Private Network Design
          Have you had success with this yet? We're trying a similar interconnect config but no long term success. We have five Dell R710 nodes and we're using the four onboard Broadcom NetXtreme II (bnx2) NICs in an 802.3ad bond with src-dst-mac load balancing and stacked Cisco 3750E-24TD-SD switches. After a random period of time (in the range of a few days to a week) all four NICs in one or more of the servers will go offline and eventually the server will reboot due to loss of the interconnect. We're about to the point of giving up on the idea of an 802.3ad bond and instead use two connections in an active/standby bond. We've updated the kernel to have the latest broadcom drivers (1.9.3) and turned off TOE on the NICs but the problem persists. Now we're updating the firmware on the NICs to the latest version (5.0.6 NCSI 2.0.3) to see if that helps. If you have your interconnects working with 802.3ad bonding I'd like to hear if you were successful just so we know it is possible and somebody did it. Thanks.