I use Ixia to construct two streams Total 20Gbps 29760000 pps flow1 udp 64size small packet Send per second 10G bps 14880000 pps flow2 tcp 64size small packet Send per second 10G bps 14880000 pps ./dpdk-testpmd -l 4-22 -n 8 -- -i --rxq 19 --txq 19 --nb-cores 18 --rxd 2048 --txd 2048 --portmask 0xff set fwd rxonly start show port stats all testpmd> show port stats all ######################## NIC statistics for port 0 ######################## RX-packets: 103906391 RX-missed: 369790696 RX-bytes: 6234383466 RX-errors: 0 RX-nombuf: 0 TX-packets: 0 TX-errors: 0 TX-bytes: 0 Throughput (since last show) Rx-pps: 4205026 Rx-bps: 2018412608 Tx-pps: 0 Tx-bps: 0 ############################################################################ Recive per second 2018412608 bps 2g bps 4205026 pps 4 million pps rx_discards_phy drop per second 10 million pps [root@localhost ~]# ethtool -S enp202s0f0 |grep dis rx_discards_phy: 35892329864 tx_discards_phy: 0 rx_prio0_discards: 35892164419 rx_prio1_discards: 0 rx_prio2_discards: 0 rx_prio3_discards: 0 rx_prio4_discards: 0 rx_prio5_discards: 0 rx_prio6_discards: 0 rx_prio7_discards: 0 If both flow become TCP, rx_discards_phy will not be drop. flow1 tcp 64size small packet flow2 tcp 64size small packet testpmd> show port stats all ######################## NIC statistics for port 0 ######################## RX-packets: 7177423122 RX-missed: 369790696 RX-bytes: 430645390083 RX-errors: 0 RX-nombuf: 0 TX-packets: 0 TX-errors: 0 TX-bytes: 0 Throughput (since last show) Rx-pps: 29779180 Rx-bps: 14294006816 Tx-pps: 0 Tx-bps: 0 ############################################################################ 29779180 pps 29 million pps rx_discards_phy no drop [root@localhost ~]# ethtool -S enp202s0f0 |grep dis rx_discards_phy: 0 tx_discards_phy: 0 rx_prio0_discards: 0 rx_prio1_discards: 0 rx_prio2_discards: 0 rx_prio3_discards: 0 rx_prio4_discards: 0 rx_prio5_discards: 0 rx_prio6_discards: 0 rx_prio7_discards: 0 server dell poweredge r750 [root@localhost proc]# lscpu Architecture: x86_64 CPU op-mode(s): 32-bit, 64-bit Byte Order: Little Endian CPU(s): 64 On-line CPU(s) list: 0-63 Thread(s) per core: 1 Core(s) per socket: 32 Socket(s): 2 NUMA node(s): 2 Vendor ID: GenuineIntel CPU family: 6 Model: 106 Model name: Intel(R) Xeon(R) Platinum 8375C CPU @ 2.90GHz Stepping: 6 CPU MHz: 2900.000 BogoMIPS: 5800.00 Virtualization: VT-x L1d cache: 48K L1i cache: 32K L2 cache: 1280K L3 cache: 55296K NUMA node0 CPU(s): 0,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40,42,44,46,48,50,52,54,56,58,60,62 NUMA node1 CPU(s): 1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43,45,47,49,51,53,55,57,59,61,63 Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc aperfmperf eagerfpu pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch epb cat_l3 invpcid_single intel_pt ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local dtherm ida arat pln pts avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg avx512_vpopcntdq md_clear pconfig spec_ctrl intel_stibp flush_l1d arch_capabilities MLX ConnectX6 100G PCIE4 x16 [root@localhost ~]# ofed_info -s MLNX_OFED_LINUX-5.5-1.0.3.2: [root@localhost ~]# mlxfwmanager Querying Mellanox devices firmware ... Device #1: ---------- Device Type: ConnectX6 Part Number: MCX653106A-ECA_Ax Description: ConnectX-6 VPI adapter card; H100Gb/s (HDR100; EDR IB and 100GbE); dual-port QSFP56; PCIe3.0 x16; tall bracket; ROHS R6 PSID: MT_0000000224 PCI Device Name: 0000:ca:00.0 Base MAC: 08c0eb204e5a Versions: Current Available FW 20.32.1010 20.32.1010 PXE 3.6.0502 3.6.0502 UEFI 14.25.0017 14.25.0017 Status: Up to date
Hello, To get the best results for mixed traffic, you need to set the correct value for the CQE compress devarg. See the following documentation snippet (I suggest reading it fully in 36.5.3.2. Driver options: rxq_cqe_comp_en parameter [int] Specifying 4 as a rxq_cqe_comp_en value selects L3/L4 Header format for better compression rate in case of mixed TCP/UDP and IPv4/IPv6 traffic. CQE compression format selection requires DevX to be enabled. Please let us know the result.
Hello, use rxq_cqe_comp_en=4 performance will be better. Now there is a new problem! 3rd Gen Intel® Xeon® Scalable Processors Receive error len error checksum UDP tcp packet performance is very low! The test environment is the same as above, use Ixia to construct two streams Total 20Gbps 29760000 pps flow1 tcp 64size small packet incorrect tcp checksum Send per second 10G bps 14880000 pps flow2 tcp 64size small packet correct tcp checksum Send per second 10G bps 14880000 pps ./dpdk-testpmd -l 4-22 -n 8 -- -i --rxq 19 --txq 19 --nb-cores 18 --rxd 2048 --txd 2048 --portmask 0xff set fwd rxonly start Rx-pps: 5134559 rx_discards_phy drop ninety percent per second ! This problem only appears on 3rd Gen Intel® Xeon® Scalable Processors platform
Can you please elaborate on the motivation for such a test? Why would you like to measure bad checksum traffic performance?
The attacker will send attack traffic, which may be various wrong packets.
Can you please try to change the rx burst function? Currently, it uses vectorized. Can you add devarg of mprq_en=1 and check the result? Another option to handle this, is to add an rte_flow rule that match on csum integrity bit and does drop action. More info can be found in http://doc.dpdk.org/guides/nics/mlx5.html
After using the mprq=1 parameter, the problem remains the same. tcp ,udp Mixed flow tcp(header checksum incorrect) ,tcp or udp(header checksum correct) Mixed flow udp(header checksum incorrect) ,tcp or udp(header checksum correct) Mixed flow tcp(header length incorrect) ,tcp or udp(header length correct) Mixed flow .... Many of these combinations performance is very low. rx_discards_PHY massive packet drop. What is the reason behind it? Is it because netcard and CPU core operate CQE at the same time? Is there any way to solve it, such as adjusting what parameters?
Can you please try to change the rx burst function? I don't understand. Is it for me to modify the code of the rte_eth_rx_burst ?
I can see two options to handle this case: 1. Add an rte_flow rule that match on csum integrity bit and does drop action. More info can be found in http://doc.dpdk.org/guides/nics/mlx5.html 2. Disable checksum validation altogether in case an application doesn’t care about checksums.
(In reply to Asaf Penso from comment #8) > I can see two options to handle this case: > 1. Add an rte_flow rule that match on csum integrity bit and does drop > action. More info can be found in http://doc.dpdk.org/guides/nics/mlx5.html > 2. Disable checksum validation altogether in case an application doesn’t > care about checksums. I tried all the methods, but they didn't take effect in dpdk Disable checksum. The network card still calculates the checksum
Hi, We have continued to look into this issue, and would like to know the following results: 1. mixed traffic TCP/UDP without cqe compression (need to set devarg rxq_cqe_comp_en=0) 2. mixed traffic TCP/UDP with cqe compression 3. mixed traffic TCP/UDP with cqe compression and csum errors
Hi @Asaf, we are observing the same issue despite rxq_cqe_comp_en=4 . On a 100Gbps load we still drop at least 40% packets as xstats rx_phy_discard_packets See bug report https://bugs.dpdk.org/show_bug.cgi?id=1053
Hello, Can you please provide the 3 results mentioned in comment #10?
Hi @Asaf It's been a year, are there new network card hardware or new (OFED/FW) solutions to the bad csum performance issue?