General recommendations for load balancing NICs: NIC #1: ZM1 - Populated mostly during daytime hours (high usage) ZE - Populated mostly during night time hours (mid usage) HS - Populated mostly during night time hours (mid usage) NIC #2: ZM2 - Populated mostly during early morning/afternoon hours (very high usage) GG - Varying use throughout the day (mid usage) CM - Varies (low usage) MG - Varies (mid usage) This should insure load balancing the distribution of network traffic. I am assuming that you have already dedicated processor affinity (dedicated) settings for each server process. If you have not already done so, you should pair the host server app with a single core. Since Xenon processors have HT (Hyper-Threading) enabled, they should be adjacent with each core (i.e. thread 0, 1 = physical processor core 0, 2, 3 = physical processor core 1). In this example, 0 and 1 should be assigned to ZM1, 2 and 3 to ZM2, etc... With a total of 24 virtual processors and 12 cores, paring is important not to combine a physical process with a virtual one. If you assign virtual process 0 to ZM1 and virtual process 1 to ZM2, you can just imagine the performance degradation with this combination... I hope this helps.