Home > Communication Service Provider Solutions > Telecom Multicloud Foundation > Red Hat > Guides > Red Hat Open Stack Platform Guides > Jumbo frames Performance Analysis Over Red Hat OpenStack 16.1 Infrastructure > Appendix B
This section highlights the troubleshooting steps carried out in the Standalone environment in collaboration with Intel.
Table 5 represents the results for the test cases mentioned in Troubleshooting section. The details of the test cases are given below.
Table 5. RFC2544 ZPL benchmark results
Test Case ID |
Test Case |
VLAN Translation |
Bonding |
RXQs, TXQs |
Line Rate (%) / PPS for 4096 Byte Packets |
Line Rate (%) / PPS for 8192 Byte Packets |
Line Rate (%) / PPS for 9000 Byte Packets |
1 |
Baremetal PF |
#N/A |
Disabled |
2 |
100 / 759,231 |
100 / 380,540 |
100 / 346,452 |
2 |
VM PF |
#N/A |
Disabled |
2 |
100 / 759,231 |
100 / 380,540 |
100 / 346,452 |
3 |
OVS-DPDK Single Bridge |
Disabled |
Disabled |
2 |
100 / 759,230 |
100 / 380,539 |
100 / 346,451 |
4 |
OVS-DPDK Dual Bridge |
Disabled |
Disabled |
2 |
100 / 759,228 |
100 / 380,539 |
100 / 346,451 |
5 |
OVS-DPDK Dual Bridge |
Enabled |
Disabled |
1 |
86.852 / 659,404 |
99.227 / 377,596 |
100 / 346,450 |
6 |
OVS-DPDK Dual Bridge |
Enabled |
Enabled |
1 |
96.906 / 735,740 |
100 / 380,540 |
100 / 346,450 |
7 |
OVS-DPDK Dual Bridge |
Enabled |
Disabled |
2 |
100 / 759,224 |
100 / 380,539 |
100 / 346,451 |
8 |
OVS-DPDK Dual Bridge |
Enabled |
Enabled |
2 |
100 / 759,232 |
100 / 380,537 |
100 / 346,452 |
Note: Unless otherwise stated, all configurations use DPDK 19.11.6 on the host/guest, with OVS 2.13.3, along with Intel® Ethernet Network Adapters XXV710-DA2, testpmd, with a 4K descriptor size, benchmarked with RFC2544 with Zero Packet Loss (ZPL).
The following test cases were run in the isolated KVM based environment (listed in order of priority) with increasingly complexity step by step to reproduce the issue.
The following run test cases define the impact of single queue for transmit and receive:
By implementing different optimizations, line rate network throughput was ensured for each test case with increasing level of virtual overhead. Test cases 1-4 with minimal infrastructure complexity reported no packet loss.
Identical test setup was simulated in test cases 5-8 with the OpenStack environment. Configuring single RX queue resulted in packet loss for all jumbo frames. However, re-running the same test setup with multiple RX queues, i.e. test case 7 and 8 eliminated the packet loss and 100% line-rate throughput was achieved.
The following figure represents isolated test setup used in test cases 5-8.
Figure 4. Isolated test setup replicating OpenStack environment