While trying to understand the DDIO misses to hit ratio, I stumbled upon the pcm-pcie.x tool.
I ran it with the as pcm-pcie.x -e -B 30 to get the extended output with DDIO, in bytes, once 30s.
There are two Xeon Gold 6126 CPU @ 2.60GHz CPUs and two X710 cards (each on a separate NUMA node).
This is a packet sniffer, AKA this host does not transfer anything into the network, it just captures packets. There's also little traffic - 1-2Gbit/sec on average.
Skt | PCIeRdCur | RFO | CRd | DRd | ItoM | PRd | WiL | PCIe Rd (B) | PCIe Wr (B) 0 50 M 28 M 26 M 26 M 16 M 21 M 16 M 8459 M 2843 M (Total) 0 33 M 31 M 28 M 21 M 14 M 20 M 21 M 7403 M 2989 M (Miss) 0 16 M 0 0 5181 K 1486 K 826 K 0 1415 M 95 M (Hit) 1 35 M 32 M 39 M 37 M 28 M 30 M 20 M 9329 M 3890 M (Total) 1 35 M 43 M 37 M 32 M 25 M 24 M 22 M 9587 M 4429 M (Miss) 1 175 K 0 1940 K 4919 K 2652 K 6050 K 0 450 M 169 M (Hit)
Link Copied
For more complete information about compiler optimizations, see our Optimization Notice.