• State TI Thinks Resolved
  • Locked Locked
  • Replies 3 replies
  • Answers 2 answers
  • Subscribers 101 subscribers
  • Views 280 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Keystone2 10Gig switch

Joey Landmark
Prodigy 110 points
Other Parts Discussed in Thread: 66AK2H14

Hi,

Is the 10Gig switch on the Keystone2 devices (66AK2H14)  a cut-through switch or a store-and-forward switch?

What is the expected switching latency ?

Thanks

  • 0
    TI__Guru*** 139515 points

    Joey

     It is a store and forward switch, the latency is approximately 250nS for a 64 byte frame from EOF on the receive port to EOF on the transmit port. And approximately 1680nS for a 2K byte frame.

    Thanks

    David

  • 0 in reply to David (ASIC) Liu
    Prodigy 110 points

    Thanks David.

    Is the latency the same if calculating for a packet received into the port 0/1 and transmitted to the Host port, or is it less since the Host port is different?

    Also, would the latency be improved if the switch is operated in ALE  bypass mode, for packets received at ports 0/1 and sent to the host port?

    Joey

  • 0 in reply to Joey Landmark
    TI__Guru*** 139515 points

    Joey

    The latency from and to the host port are lower due to the PSI bandwidth. This PSI bandwidth can transfer data as high as 40+Gbps in each direction. But since this bandwidth is dependent on the target memory and system load, latencies can vary between 260nS to 1100nS for a 2K frame.

    Although a 64 byte frame could be transferred in as little as 15nS, the per frame descriptor work and LUT work caps the per frame lower latency to about 210nS.

    For the larger frames the LUT work was buried during the transfer latency and hence does not impact larger frame latencies.

    Since the CPSW can generate 14.88 million packets per second per port per direction for 64 byte packets, the SW driver processing efficiency is of paramount importance. That is if the SW can only transmit say 7 million packets per second, then the SW will not hit transmit wire rate until 160 byte packets for a single port or 344 byte packets for both ports assuming no receive impact.

    Thanks

    David