[KubeCon18] Benchmark 3 Service Densities of the Snake test case on KVM

[taylor]

Run benchmark tests and add the results to this ticket. All tests will be run against a machine running KVM.

Use the same number of cores and threads which are used for the K8s benchmark tests. See #159

• [x] Document KVM snake test case in this ticket

Quad Intel Snake Test

• [x] Deploy KVM system on a quad intel m2.xlarge Packet system
• [x] Setup VPP vSwitch similar to K8s worker node
• [ ] Deploy VNFs to KVM in snake test cases:
  • [ ] KVM Intel Snake (VSC)
    • [ ] 3 Chains, 1 NF
    • [ ] 3 Chains, 2 NFs
    • [x] 2 Chains, 3 NFs
• [x] Run nfvbench against KVM system for each service density and test case above
• [ ] Collect results and save logs to this ticket
• [ ] Summarize results, including a single throughput number
• [ ] Add system resource information, including memory and cpu usage by VNFs and KVM to this ticket

Dual mellanox Snake test

• [x] Deploy KVM system on a dual mellanox m2.xlarge Packet system
• [x] Setup VPP vSwitch similar to K8s worker node
• [x] Deploy VNFs to KVM in snake test cases:
  • [x] KVM Mellanox Snake (VSC)
    • [x] 3 Chains, 1 NF
    • [x] 3 Chains, 2 NFs
    • [x] 2 Chains, 3 NFs
• [x] Collect results and save logs to this ticket
• [x] Summarize results, including a single throughput number
• [ ] Add system resource information, including memory and cpu usage by VNFs and KVM to this ticket

[taylor]

NFVbench results for VSC testing on Intel

• 2 Chain 3 network functions:

  • 4t2c_novlan/vsc_2c3n_18500000pps-1.log
  • PPS for 64B packets 909,840 pps (611.4133 Mbps)

• 3 Chain 2 network functions:

  • 4t2c_novlan/vsc_3c2n_18500000pps-1.log
  • PPS for 64B packets 1,105,058 pps (742.5996 Mbps)

[michaelspedersen]

Summary of "VSC KVM results using MLX (Packet.net)"

Setup:

• 4t2c host VPP
• 1 RX queue per vHost (2 per VNF)
• m2.xlarge.x86 server.

Test:

• MRR (10Gbps)
• 64B packets
• 60 second duration
• Single iteration

Values in the matrix below are MPPS

VMs, MLX (Packet.net), VSC
VSC   001   002   003   004   006
001   2.70* 2.56  1.89* 1.77  1.40
002   4.96  3.23  2.57  ----  ----
003   4.93* 3.18  ----  ----  ----
004   4.87  ----  ----  ----  ----
006   4.77  ----  ----  ----  ----

(*) Indicates non-optimal (uneven) distribution of vHost queues in host VPP, which may impact performance.

List of results (MPPS and Gbps):

vnf_mlx_1c1n_10Gbps-1.log: 2,699,749 pps (1.8142 Gbps)
vnf_mlx_1c2n_10Gbps-1.log: 2,560,264 pps (1.7205 Gbps)
vnf_mlx_1c3n_10Gbps-1.log: 1,887,053 pps (1.2681 Gbps)
vnf_mlx_1c4n_10Gbps-1.log: 1,771,559 pps (1.1905 Gbps)
vnf_mlx_1c6n_10Gbps-1.log: 1,399,602 pps (940.5337 Mbps)
vnf_mlx_2c1n_10Gbps-1.log: 4,964,925 pps (3.3364 Gbps)
vnf_mlx_2c2n_10Gbps-1.log: 3,230,608 pps (2.1710 Gbps)
vnf_mlx_2c3n_10Gbps-1.log: 2,573,710 pps (1.7295 Gbps)
vnf_mlx_3c1n_10Gbps-1.log: 4,929,237 pps (3.3124 Gbps)
vnf_mlx_3c2n_10Gbps-1.log: 3,184,974 pps (2.1403 Gbps)
vnf_mlx_4c1n_10Gbps-1.log: 4,873,372 pps (3.2749 Gbps)
vnf_mlx_6c1n_10Gbps-1.log: 4,771,932 pps (3.2067 Gbps)