Skip to end of metadata
Go to start of metadata
Icon

All performance figures are as measured on ICO-TTJ-500 using v2.0.3 BSP which is based on a Freescale IMX6q processor (1GHz) running Linux with RT patches and SDIO driver optimizations.

Loading any network filtering, NAT or IP masquerading kernel modules may result in lower performance numbers owing to increased IP packet processing overhead. Remove all unnecessary modules using rmmod.

Performance results on ICO-TTJ-500

TestMin.Typ.Max.Notes
Connection establish latency 300 us Measured directly over CNL from initiator CNLIO_connect request to CNLIO_EVT_ACCEPT_IND.
Data transmission latency (power saving off) 250 us Measured directly over CNL from initiator CNLIO_sendData to CNLIO_EVT_REQ_COMP for a small packet when power saving is off.
Data transmission latency (power saving on) 0.5 s Measured directly over CNL from initiator CNLIO_sendData to CNLIO_EVT_REQ_COMP for a small packet when power saving is on.
CNL peak unidirectional throughput (UHS-I mode)335 Mbps340 Mbps347 MbpsMeasured directly over CNL with 256K CNL buffer size.
Obex peak unidirectional throughput255 Mbps260 Mbps265 MbpsMeasured using Icoteq demo application with GUI.
TCP/IP unidirectional throughput139 Mbps150 Mbps157 MbpsMeasured using IP tunnelling application (directly over CNL) with 256K CNL buffer size.
TCP/IP bidirectional throughput 75 Mbps Measured using IP tunnelling application (directly over CNL) with 256K CNL buffer size in both directions with alternating transfer direction.
UDP throughput  176 Mbps

Measured using IP tunnelling application (directly over CNL) with 256K CNL buffer size.

This is the maximum achievable with 0% packet loss.

Ping round-trip delay815 us1507 us3663 usMeasured using IP tunnelling application (directly over CNL) with power saving off.
CNL peak unidirectional throughput (HS mode) 136 Mbps Measured directly over CNL with 256K CNL buffer size.

 

Observations

Where data transmission latency is critical, the power saving feature can be circumvented by sending dummy "keep alive" packets over the link to prevent power saving mode from being entered.  Otherwise, a one-time penalty of 0.5s will be incurred when power saving mode is active.

The peak CNL throughput depends on the SDIO operating mode i.e., HS mode or UHS-I mode.  The only difference is the SDIO bus clock speed.  A higher SDIO bus clock speed reduces the inter-packet gap on contiguous transmissions.

The CNL medium sensing collision avoidance scheme does not provide a robust method for managing bi-directional communications.  It is not advisable to treat the CNL as a full duplex pipe since it will lead to data collisions which reduces overall throughput.  Where a full duplex operation is needed, this is best managed by the application protocol which can determine when to grant transmission opportunities over the CNL link.

This concept is exploited in the IP tunnelling application by allowing initiator and responder equal opportunities to transmit data by regularly turning the link direction around.  However, this reduces peak throughput by virtue of the overhead in turning the link direction around.  This can be seen by the fact that UDP maximum throughput is only 170 Mbps versus a maximum of 324 Mbps when the transmission link is not turned around.