On the slave side of the connection, you can determine the data flow direction for the connection. If the data flow is to be one-way, from the master to the slave (i.e. the slave will only read from the master, not write), you will get the fastest performance by configuring the connection "read-only", as follows:
On the DataHub instance; that is making the slave side connection, right click on the DataHub system-tray icon and choose .
In the Properties window, select Tunnel/Mirror. 
In the Data Flow Direction section, select .

Click .
Ensure that the slave side of the tunnel/mirror connection is set to use binary mode transmission:
On the DataHub instance that is making the slave side connection, right click on the DataHub system-tray icon and choose .
In the Properties window, select Tunnel/Mirror. 
In the Tunnel/Mirror Slave section, highlight the host name of the tunnelling master, and click the button to open the Tunnel/Mirror Master Configuration window.
Check the box.

Click .
For more information, please refer to Section 22.1, “Binary Mode Tunnel/Mirror (TCP) Connections”.
To optimize performance for the network heartbeat and timeout, please refer to Section 22.2, “Tunnel/Mirror (TCP) Heartbeat and Timeout”.
Configure the master side of the tunnel/mirror connection:
On the DataHub instance that is the master for the tunnel/mirror connection, right click on the DataHub system-tray icon and choose .
In the Properties window, select Tunnel/Mirror. 
In the Tunnel/Mirror Master section, you can choose between one of three states:
No queuing with buffered delivery This is the fastest state. To do this, un-check the option in the Tunnel/Mirror configuration tab:

Queuing with buffered delivery This is a reasonable compromise that will keep up to three old values for each data point if there is a short burst of heavy data traffic. If there are more than three values for a point outstanding, the oldest will be discarded. To do this, check the option , but do not select any of the data types below it:

Queuing with un-buffered delivery This is the slowest mode, but also the one least likely to discard old values during short periods of heavy data traffic. In this mode there is a queue of up to 3 old values for each data point. In addition, you may choose to force the data transmission when a point with any of the specified types changes. The goal here is to possibly allow buffering (and hence, possible discarding of old values) for some types, while attempting to preserve all changes for other types. For example, here we have chosen to force an outbound transmission if any boolean or string value changes, but to buffer any changes in floating point and integer types:

Click .
For more information, please refer to Section 22.3, “Old Value Queuing” and Section 22.4, “Un-Buffered Delivery”.