2545 - LatticeECP3: Why does the SGMII and Gb Ethernet PCS IP reduce the Inter Packet Gap between Ethernet frames?
Description:The Inter Packet Gap (IPG) shrink (deleting of /I2/ ordered set) will always happens as long as there is a PPM (Part per Million) difference between the local reference clock and the remote reference clock, and the remote reference clock is faster than the local reference clock .
Increasing the IPG between frames does not prevent the IPG shrink. For example, when you increase the IPG from 12 bytes to 14 bytes, the CTC may still shrink the 14 bytes to 12 bytes or even 10 bytes with very large frame sizes:
In practice, when the GMII data rate is 1Gbps, and the maximum frame size is 1500 bytes, you should not expect more than one idle code group (2 bytes) to be deleted during the IPG period. The expected maximum clock slip for +/- 100 PPM offset is about M=0.3 clock periods per frame.
However, for Jumbo frames (let's say larger than 15,000 bytes ) the clock slip per frame may be larger than M=2.0 clocks per frame. In these cases, you can expect up to two consecutive idle code groups dropped (4 bytes of IPG).
In summary, you can expect IPG shrink of 2 bytes on normal frames and IPG shrink of 4 bytes on large jumbo frames (> 15000 bytes). For more information on how we compute M, you can see the "Guidelines for Calculating Static CTC FIFO Thresholds" section of IPUG60.
Note that if the local reference clock is faster then the remote reference clock, then you should expect the IPG to increase instead of shrinking. This information is documented in IPUG60.