Traffic flows for Cisco UCS & HP Virtual Connect

[SYSTEM]

Inside an Enclosure where the blades are connected to different sides

UCS.jpg

VC.jpg

UCS operating in “End-Host” mode

The Cisco UCS iN comparison is end host mode.

“End Host Mode”, does not behave like a normal Layer 2 switch on its uplinks.  
Instead, server NICs are “pinned” to a specific uplink, and no local switching happens from uplink to uplink.  This allows “End Host Mode” to attach to the network like a “Host” without spanning tree, and all uplinks forwarding on all VLANs. End Host Mode is the preferred mode, and it’s enabled by default.

Here we compare communication between VMs when the are attached to DIFFERENT sides. For Cisco UCS the traffic flow is from the VM to the FEX, along an external cable to the Fabric Interconnect but because the Fabric Interconnect do not pass data traffic between them the path must traverse to a switch above the Fabric Internconnect and then back.

The CALCULATED latency using the specs from Cisco is 7.4 microseconds. By statistics half of all intra server traffic needs to be forward up to this point. Compare this to Virtual Conmnect where the different side communications is within the Enclosure between two VC modules had has a CALCULATED latency of 3 microseconds. This is 59% lower latency with HP.


Latency Substantiation:
SFP+ CU copper cable 0.1us
http://www.cisco.com/en/US/docs/unified ... rview.html

UCS 6200  3.2 to 2usec
http://www.cisco.com/en/US/products/ps11548/index.html

Nexus5548   2usec
http://www.cisco.com/en/US/prod/collate ... 22479.html              

UCS-IOM-2208XP 0.6usec 
http://www.cisco.com/web/FI/expo2011/pr ... _Salli.pdf
UCS-IOM-2204XP 0.5 usec

Virtual Connect Flex Fabric 1.5 µs Ethernet ports and 2.0 µs Enet/Fibre Channel ports
http://h18004.www1.hp.com/products/quic ... 652_na.pdf