Radisys 7010 NPU Routing for Shared Metarouter Cards

1
Radisys 7010 NPU Routing for Shared Metarouter
Cards
David M. Zardzar_at_wustl.edu http//www.arl.wustl.
edu/arl
2
Radisys 7010 Shared Metarouter Card

• Physical Problem We wish to route packets coming
  into a metarouter (MR) card to the NPU where the
  MR lives.
• Logical Problem In the general case, where
  packets are arriving over the fabric switch on a
  10 Gb/s link to the FIC, we cannot inspect packet
  data to route packets based on VLAN (or anything
  else).

3
The FIC

• It has been reported that the FIC has an Intel
  IXF18203 Ethernet MAC to SPI-4 device on it.

4
The FIC Continued

• Now each channel can be routed through the SPI-4
  switch to the proper NPU.

5
Addressing MR/NPUs

• To take advantage of this scheme
• each MR board must be connected to both switches
  in order to use both NPUs.
• There is no failover for if one switch goes down,
  there is no way to get packets to the NPU
  accessed by that switch
• The MAC address of one of the IXF18203 MAC
  devices is the logical MAC address of the NPU.

6
Lookup Scheme

• Packets arriving at the LC need to know where to
  go
• When a lookup is done to determine which MR is
  the destination, the logical MAC address of that
  MR needs to be inserted into the Enet header (no
  different than before except there are two MAC
  addresses per shared- MR card), and
• The SPI-4 channel number that is associated with
  that MAC address is sent to the TX block so that
  it can send the packet to the correct switch that
  can see this MAC address.
• The reverse path works the same way
• When a lookup is done to send data from a MR to
  somewhere else, if that somewhere is another MR,
  it needs to know the logical MAC address of that
  MR and insert that into the Enet header.
• At the same lookup, the SPI-4 channel number to
  send out on is also extracted and passed to the
  transmit block so it is transmitted on the
  correct channel to reach a switch that can see
  the specified MAC address.
• If the destination is an LC, we could use either
  channel number as its SPI-4 switch would be
  configured to send all traffic to one NPU (the
  ingress NPU). If we keep the same channel as the
  incoming packet, however, then we know it will go
  out through the same switch that sent the packet
  to the MR.

7
Future Scheme

• Radisys has mentioned that the next version of
  the dual NPU blade will have the FIC on the board
  in the form of an FPGA that we will be able to
  reprogram.
• We could then inspect the packets for VLAN tag to
  determine which SPI-4 channel to use.
• We could have full failover, again, as all
  traffic could go to either switch and still be
  routed to the correct NPU through the FPGA
  lookup.
• The timing of this new board is sometime in
  2007.