B.2. Step Two: Does TCP optimization and data reduction work?

B.2.1. Issue one: The inner channel doesn't stay up.

As long as the client and the server don't close the TCP session, the inner channel should stay up.

  • The TCP connection times out.

    • The inner channel doesn't support fragmentation. If there is a gateway in the path which doesn't support for the MTU size of the in-path interface, we expect to see an ICMP Fragmentation Required packet with the right MTU size. If that packet doesn't get delivered, the TCP session will time out.

  • The TCP session gets reset.

    • A firewall in between the CSH and SSH has noted that the final ACK of the setup of the original TCP session didn't arrive. After a timeout the firewall will send a TCP RST to the client and server, intercepted by the CSH and SSH which will terminate the TCP session. This will affect long-lived TCP sessions most.

    • With the Full Transparency or Port Transparency WAN visibility, an IPS/IDS will find out that the protocol spoken on the inner channel isn't the protocol expected and it will terminate the TCP session.

B.2.2. Issue two: Traffic is not suited for optimization.

The traffic must be suitable for optimization.

  • Bad data reduction.

    • Encrypted traffic. If it is SSL encrypted it could be optimized with SSL pre-optimization. If anything else, no luck. Check the documentation of the application to see if this can be disabled.

    • Compressed traffic. Check the documentation of the application if this can be disabled.

  • Bad traffic scenario.

    • If the traffic between the client and server is not a stream of lots of traffic in burst but more a ping-pong, the neural framing algorithm will cause extra delay here.

B.2.3. Issue three: Network issues

If the network isn't happy, the optimized traffic won't perform well neither.

  • Speed / duplex mismatch.

  • Congestion.

    • QoS not enabled.

    • QoS bandwidth misconfigured.

  • Large Fat Network.

    • WAN link doesn't get filled. Use HS-TCP or MX-TCP.

  • WAN bandwidth admission control.

B.2.4. Issue four: Steelhead related issues

And it could just be that the SH is not performing well.

  • CPU load, paging.

  • Broken disk.

  • High disk utilization: SDR Cost / SDR Latency / Disk Load 3.

  • QoS configuration wrong.

  • Optimization service not happy.