4 Replies Latest reply on Nov 5, 2017 7:35 PM by Budweiser

    x550-t2のpps処理の限界について

    Budweiser

      質問をしますので、詳細をご存知の方、ご回答いただければ幸いです。サーバ1(送信側)、2(受信側):NIC x550-xt DPDK 実装  サーバ1で、pktgenを起動して64byteのパケットを1480万ppsでサーバ2に送信しましたが、サーバ2では、480万ppsほどしか記録が出ません。色々試しましたが、480万がx550-t2で処理できるパケット数の上限値と考えてよいのでしょうか。もし、この数値が上がる方法があったらご教示願います。

        • 1. Re: x550-t2のpps処理の限界について
          N.Scott.Pearson

          Hhmmm, doing the math, 4.8M * 64B * 8b == 2.29Gb transfer rate. How much higher were you expecting?

          • 2. Re: x550-t2のpps処理の限界について
            Budweiser

            Thank you for your reply. I was expecting 10gbps because It was 10 Gbps when I tested x 540 in the same environment. I used 14880000 packets with pktgen. As you know, if it's a 64 byte packet, it is equivalent to 10 G with 14880000 packets.

            In the measurement, x540 - t2 received all 14.88 million packets, while x550 - t2 received only 4.8 million packets.

            As for specs, the expected result of x550 - t2 was worse than x540 - t2, although the performance is better for x550 - t2 than for x540 - t2. If you know the reason why it will result like this, could you teach me ?

            • 3. Re: x550-t2のpps処理の限界について
              N.Scott.Pearson

              What I was getting at is that there can be transactional costs that limit the overall effective bandwidth. If you were to, for example, send 256 byte packets, I am sure you would see the effective bandwidth improve.

               

              Yes, it would seem that the performance of this adapter is worse than the previous. We need folks with the right knowledge level to answer this question. I am going to move this conversation into the Wired Ethernet forum, where more-knowledgeable folks should be lurking...

               

              ...S

              • 4. Re: x550-t2のpps処理の限界について
                Budweiser

                To the respondent

                 

                Thank you for your valuable answer.I think your response is credible.

                because our measurement results match what you are saying.

                By moving this discussion to Wired Ethernet forums, I will expect high precision answers

                and maybe ask someone who is familiar with this matter like you again in that forum. Thank you very much and I look forward to your kind cooperation.