Offloading the OSPF Protocol: HELLO Processing


Current network device implementations using the IXP1200 have fast-path
packet forwarding occurring completely on the microengines (forwarding
plane) and signaling protocols such as OSPF on a separate x86 processor
(control plane).  This design presents several weaknesses, for example,
the single control plane constitutes a central point of failure and a
bottleneck.  One possible preliminary solution for these problems is to
offload distinct and repetitive operations of control plane routing
protocols to the data plane line cards.  The OSPF HELLO algorithm is a
good candidate for such offloading, as it employs soft state mechanisms
that require periodic update messages to be transmitted, received and
processed.  As a first step toward this goal, my summer project focused on
splitting OSPF HELLO processing from a centralized OSPF signaling state
machine and offloading it to the StrongARM processor available on the
IXP1200.  If offloading such protocols turns out to be successful, the
flexible and programmable design of the Intel Network Processors could use
these offloading techniques to architect and implement network elements
that are scalable and very responsive in the the presence of failures.