Troubleshooting
VXLAN Speed
EgressGateway uses the vxlan tunnel, and testing shows that vxlan loss is around 10%. If you find that the speed of EgressGateway does not meet the standard, you can follow these steps to check:
- Confirm that the speed of the host-to-node matches the expected speed;
- The offload setting of the network card used by vxlan on the host will have a small impact on the speed of the vxlan interface (there will only be a difference of 0.5 Gbits/sec in the 10G network card test), you can run
ethtool --offload host-interface-name rx on tx on
to turn on offload;
- The offload setting of the vxlan network card can significantly impact the speed of the vxlan interface. In 10G network card tests, the speed is 2.5 Gbits/sec without offload enabled, and 8.9 Gbits/sec with offload enabled. You can run
ethtool -k egress.vxlan
to check whether checksum offload is turned off, and you can enable offload by setting the feature.vxlan.disableChecksumOffload
configuration in helm values to false
.
Benchmark
The following are the data from our stress tests using bare metal servers.
Name |
CPU |
MEM |
Interface |
Node 1 |
Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz |
128G |
10G Mellanox |
Node 2 |
Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz |
128G |
10G Mellanox |
Node Target |
Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz |
128G |
10G Mellanox |
Case |
Item |
Detail |
case1 |
node -> node |
9.44 Gbits/sec sender - 9.41 Gbits/sec receiver |
case2 |
egress vxlan -> egress vxlan |
9.11 Gbits/sec sender - 9.09 Gbits/sec receiver |
case3 |
pod -> egress node -> target |
9.01 Gbits/sec sender - 8.98 Gbits/sec receiver |
Virtual machine
The following is a virtual machine using VMWare, which limits the data measured with Node specification of 4C8G.
Name |
CPU |
MEM |
Interface |
Node 1 |
Intel(R) Xeon(R) Gold 5118 CPU @ 2.30GHz 4C |
8G |
VMXNET3 |
Node 2 |
Intel(R) Xeon(R) Gold 5118 CPU @ 2.30GHz 4C |
8G |
VMXNET3 |
Node Target |
Intel(R) Xeon(R) Gold 5118 CPU @ 2.30GHz 4C |
8G |
VMXNET3 |
Case |
Item |
Detail |
case1 |
node -> node |
2.99 Gbits/sec sender - 2.99 Gbits/sec receiver |
case2 |
egress vxlan -> egress vxlan |
1.73 Gbits/sec sender - 1.71 Gbits/sec receiver |
case3 |
pod -> egress node -> target |
1.23 Gbits/sec sender - 1.22 Gbits/sec receiver |