Performance

Latency times much better than for TCP

  • ~33% faster than TCP inter node
  • 2 times faster than TCP intra node for 64 byte messages
  • 7 times faster than TCP intra node for 64 kB messages
    -> TIPC transmits socket-to-socket instead of via loopback interface


Throughput lower than TCP on bare metal and between VMs

  • ~65-90% of max TCP throughput inter-node
    -> Environment dependent, notably due to TCP TSO
    -> We are working on this

Throughput much better than TCP when intra-node/host

  • 4 times better than TCP
    -> New “smart nagle ” feature improves small message throughput
  • Turns itself on an off depending on traffic pattern
    -> No ACK delay problem like in TCP NAGLE


"Wormhole" technique shortcuts messaging between kernel name spaces

  • Same latency and througput as for intra-node traffic
  • Messages passed directly between sockets
  • No need to pass through interfaces
  • Peer authentication, but no need for encryption
  • Useful for Kubernetes pods