- Integrated Systems
- About Us
- Integrated Systems
- About Us
02-14-2011 07:22 AM
Internal chassis blade communications with Virtual Connect
Stephen was looking for help on Virtual Connect for a customer situation:
> Hi - customer wants to create two separate SUSes on the FlexFabric
> modules in rear bays 1 and 2 so that they operate in active/active
> mode to give the full uplink bandwidth.
> If they do this they have asked the following 3 questions:-
> (1) If blades within the chassis need to communicate via the SAME port
> (e.g. port 1 or port 2) does traffic go out to the upstream switches
> or do the VC modules route traffic for the same chassis directly back
> down the relevant downlink? (I'm pretty confident that the answer is
> YES but I need confirmation).
> (2) if two blades within the chassis need to communicate via DIFFERENT
> NIC ports (e.g. Blade-X via sends traffic via port 1 to port-2 on
> Blade-Y in the same chassis) will this traffic be routed/bridged
> across the X7,X8 cross-links OR will it need to travel out of the VC
> uplinks to the external switches and then be sent back to BladeY. So
> they will have created, say, VLAN101-1 on SUS1 and VLAN101-2 on SUS2
> and ideally would prefer these two networks to communicate across the
> cross-links rather than be routed via the external switches. I suspect
> the answer to this is NO because there is no communication between
> different VLANs across the interconnect? Can anyone confirm or deny?
> (3) Today they use Cisco blade switches. For VMWare ESX, they tell me
> all network traffic from any individual VM can only be routed out
> through one port on the host blade (I guess because they cannot create
> an active-active trunk across two Cisco blade switches (unless they
> use the Cisco VBS feature which they don't, as it has other issues).
> They can manually load balance different VMs across the two blade
> ports but would prefer to create a single aggregated link from the two
> 10G ports on the BL460cG7 and give every VM access to the full 20G
> uplink bandwidth (or some Flex-10 fraction of that). I know that
> aggregation of the two NICs on the BL460cG7 can only be done by the
> operating system in the case of Windows and I assume that ESX 4.x can
> do the same but I need confirmation. Does ESX 4.x allow creation of a
> single aggregated 20G NIC on the BL460cG7? Again I think answer is YES
> but I need confirmation.
Here some rules how traffic is forwarded with VC:
- If traffic stays in the same SUS it will switched locally and use the cross links (in a 2 interconnect module design)
- Traffic between 2 different SUS need always be switched by the next uplink switch
- Load Balancing on ESX can have different flavors with VC
Route based on the originating vSwitch port ID <- recommended for most setups
Route based on source MAC hash
Route based on IP hash
VC does not support LACP on downlinks
VC does never route. It just forwards a packet. Just to avoid the wrong terms.
Hope this helps already.
Other comments or suggestions for Stephen?