Got here a bit late as stayed to talk with Bryan Kuhn and Chad Sakac after the last session….discussion around NFS for VDI, etc.

Speaker = Don Mann from ePlus

• Two Approaches
  • Throttling – examples…
    • vSS – throttle egress (outbound) traffic from VM
    • vDS – throttle ingress/egress to/from VM
    • HP FlexFabric – FlexNICs – 4 NICs per 10 Gb link
  • Prioritization
    • vDS with vSphere 4.1 or higher – Network I/O Control
    • Cisco Nexus 1000v — as of 1.4 can do Bandwidth Prioritization Queueing
    • Cisco Virtual Interface Card – part of UCS – very clean.
• vSwitch Options in vSphere 5 – some of these are new
  • vNetwork Standard Switch – same as vSwitch in VI3
  • Distributed vSwitch — private VLANs, NIOC, Datacenter QoS, DVMirror, NetFlow
  • Nexus 1000v — network managed by net team, “Virtual Security Gateway”, Nexus 1010 w/Network Analysis Module
• Not just the vSphere hosts….
  • Total Solution Design
  • Traffic congestion possible at…
    • blade chassis uplinks
    • storage uplinks
    • switching interconnects
  • Link Failure Detection
    • Cisco Link State Tracking
    • ESX Beacon Probing
    • HP SmartLink
• Quality of Service
  • Priority Flow Control 
    • enables lossless Fabrics for each class of service.
    • PAUSE sent per virtual lane when buffers limit exceeded.
  • Class of Service based Bandwidth Management
    • 802.1Qaz for enhanced.
  • FCoE — biggest difference is the physical media.
    • FC frames over L2 Ethernet transport – lossless stuff in 10 GigE standard to enable this
    • Single adapter
    • No gateways
    • Mostly single hop..a bit of multi-hop but not much.
    • FCoE doesn’t require SAN support….can do FCoE to hosts and FC to SAN
• Very good network stack comparison slide..get this from the slide deck.
  • Less overhead than FCIP or iSCSI
• General recommendations
  • read vendor best practices
  • Check HCL and update drivers/firmware as needed.
  • Indentify network failures, Beacon Probing, SmartLinks, Link State Tracking
  • Utilize vmxnet3 where possible
  • Leverage Twinax cables where feasible < 10m (MUCH lower cost)
  • Jumbo Frames — should only be enabled for unique workloads
  • Test, Test, Test
    • Ideally before it goes into production
    • Verify the throughput/functionality on the card.
  • Enable spanning-tree port fast on upstream switch ports.
  • VLAN tag into vSphere — provides easier troubleshooting
  • Consider Multi-NIC vMotion with proper traffic management.
• Rackmount server recommendations
  • Leverage Network I/O Control features
    • Enable QoS priority tag (and on upstream switch)
      • Or throttle vMotion at ~3-4 Gbps (on vSS and vDS)
    • Enable NetIOC on 6Gbps if leveraging FCoE (4 GB -> FC)
  • Optionally — upgrade to 1000v to gain:
• HP FlexFabric Recommendations — multiple recommendations…check the slides.
• Cisco UCS Recommendations
  • Leverage Cisco VIC – Hardware QoS on adapter
  • Fully configure QoS
    • Many best practice docs are missing some of the steps.
  • Leverage port-channels and pin groups for uplinks
  • Consider Updating Templates for consistency
  • Set Maintenance Policy to “user-ack”
  • Enable CDP under Network Control Policy
• 3 Ways to use Cisco VIC card
  • Adapter FEX — likes this one the most…..simplest as shows vmnics in the same way to ESX as normal setups.
  • 1000v
  • VM FEX — gives vNICs directly to VMs