Network performance testing, especially in 5G network topology, is a highly complicated and specialized expertise in the IT industry. As network infrastructure becomes increasingly cloudified into software-defined networking (SDN) and network functions virtualization (NFV) popularized, optimization of network efficiency has become a hot topic discussed by IT engineering and operations teams in technical forums, especially concerning CAPEX and OPEX, as they reach for the ultimate goals of achieving simplified and fast-tracked trial testing with low-cost deployment.
Network and communication equipment developers used to acquire costly testing equipment from established brand names, such as IXIA and Ciena, incurring huge expenses not only from the initial purchase of hardware, application packages, and training but also subsequent maintenance services. This high capital investment has kept many network appliance producers from achieving network technology leadership.
But there’s another choice. By adopting automated test solutions that are based on both NEXCOM’s high-performance and reliable product lines and O’Prueba’s long-established and industry-specific expertise in real-world traffic emulation and automated network testing tools with its own ACTS (automatic control testing system), developers can easily execute comprehensive test planning and validation without traversing the long route of budgeting, approval, and purchasing.
This white paper demonstrates the ACTS testing solution based on NEXCOM’s NSA 7141 platform. As Figure 1 shows, NSA 7141 is an 1U rackmount appliance supporting a single Intel® Xeon® scalable processor and memory size up to 256GB DDR4 RDIMM. The system accommodates four LAN module expansions for a wide range of bandwidth configurations to allow simulation and testing of diversified network test cases and applications.
Figure 1. NEXCOM NSA 7141 network testing platform
As shown in Table I, the system configuration consists of NSA 7141 and NX 140F, a LAN module powered by Intel® XL710 and supporting four 10GbE SFP+ ports.
TABLE I
SYSTEM CONFIGURATION
Equipment |
Description |
Platform |
NEXCOM NSA 7141 |
CPU |
Intel® Xeon® Gold 6150 CPU @ 2.70GHz |
Total Memory |
DDR4 256GB |
Storage |
1 TB |
LAN Module |
NEXCOM NX 140F |
The topology for testing a single LAN module NX 140F is shown in Figure 2. The tester is the Spirent N4U system. This 4x 10GbE module is connected via 10G DAC cables to Spirent N4U, with packet size set to 64 bytes to generate unidirectional packet traffic. One transmit queue is allocated for each port with affinity set to unique logical cores. Table II shows the test results. Compared with the theoretical maximum throughput frame size of 64 bytes displayed in Table III, the TX rate falls between 81 to 83%. Table IV displays Intel’s limits for its Ethernet Converged Network Adapter X710-DA4’s 4x10 GbE performance with 64-byte packet sizes, and Figure 3’s zero packet loss testing results further demonstrates that 64-byte packet sizes cannot reach a 100% TX rate.
Figure 2. Single NX 140F TX throughput test (4 ports per LAN module)
TABLE II
TEST RESULTS FOR SINGLE NX140F TX THROUGHPUT TEST (4 PORTS PER LAN MODULE)
Port |
Throughput (FPS) |
TX Rate (%) |
Port 1 |
12216182 |
81.72% |
Port 2 |
12223121 |
82.14% |
Port 3 |
12231141 |
81.19% |
Port 4 |
12201234 |
81.99% |
TABLE III
THEORETICAL MAXIMUM FRAME RATES FOR DIFFERENT FRAME SIZES
Theoretical maximum Frame Rates (frames/second) for different frame size (bytes) |
|||||||
Speed |
64 |
128 |
256 |
512 |
1024 |
1280 |
1518 |
10 Mbps |
14881 |
8446 |
4529 |
2350 |
1198 |
962 |
813 |
100 Mbps |
148810 |
84460 |
45290 |
23497 |
11973 |
9616 |
8128 |
1000 Mbps |
1488096 |
844595 |
452899 |
234963 |
119732 |
96154 |
81275 |
10 Gbps |
14880952 |
8445946 |
4528986 |
2349625 |
1197318 |
961539 |
812744 |
25 Gbps |
37202380 |
21114864 |
11322463 |
5874060 |
2993295 |
2403846 |
2031859 |
40 Gbps |
59523809 |
33783783 |
18115942 |
9398496 |
4789272 |
3846153 |
3250975 |
100 Gbps |
148809523 |
84459459 |
45289855 |
23496240 |
11973180 |
9615384 |
8127438 |
TABLE IV
INTEL® TEST RESULTS FOR INTEL® ETHERNET CONVERGED NETWORK ADAPTER X710-DA4[2]
Packet Size (Bytes) |
Throughput (Mpps) |
Line Rate % |
64 |
36.5 |
61.33 |
128 |
33.78 |
100 |
256 |
18.11 |
100 |
* Note: All packet sizes over 128B reach line rate.
Figure 3. RFC2544 Zero packet loss test on
Intel® Ethernet converged network adapter X710-DA4[2]
The topology for testing dual NX 140F 10GbE modules is shown in Figure 4. The tester is the Spirent N4U system. Each NX 140F module consists of 2x 10GbE, and a total of 4x 10GbEs are interconnected via 10G DAC cables. The traffic is unidirectional with packet size of 64 bytes. In this test, each slot only uses two ports. One transmit queue is allocated for each port with affinity set to unique logical cores. Test results appear in Table V. ompared with the theoretical maximum throughput frame size of 64 bytes (Table III), the TX rate reaches 100%.
Figure 4. Dual NX 140F TX throughput test (2 ports per LAN module)
TABLE V
TEST RESULTS FOR DUAL NX 140F TX THROUGHPUT TEST (2 PORTS PER LAN MODULE)
Port |
Throughput (FPS) |
TX Rate (%) |
Slot1-Port 1 |
14880952 |
100% |
Slot1-Port 2 |
14880952 |
100% |
Slot2-Port 1 |
14880952 |
100% |
Slot2-Port 2 |
14880952 |
100% |
The topology for testing dual NX 140F 10GbE modules is shown in Figure 5. The tester is the Spirent N4U system. Each NX 140F module consists of 3x 10GbE, and a total of 6x 10GbEs are interconnected via 10G DAC cables. The traffic is unidirectional with packet size of 64 bytes. In this test, each slot only uses two ports. One transmit queue is allocated for each port with affinity set to unique logical cores. Test results appear in Table VI. Compared with the theoretical maximum throughput frame size of 64 bytes (Table III), the TX rate reaches 100%.
Figure 5. Dual NX 140F TX throughput test (3 ports per LAN module)
TABLE VI
TEST RESULTS FOR DUAL NX 140F TX THROUGHPUT TEST (3 PORTS PER LAN MODULE)
Port |
Throughput (FPS) |
TX Rate (%) |
Slot1-Port 1 |
14880952 |
100% |
Slot1-Port 2 |
14880952 |
100% |
Slot1-Port 3 |
14880952 |
100% |
Slot2-Port 1 |
14880952 |
100% |
Slot2-Port 2 |
14880952 |
100% |
Slot1-Port 3 |
14880952 |
100% |
Topology for testing multiple NX 140F 10GbE modules is shown in Figure 6. The tester is the Spirent N4U system. Each NX 140F module consists of 2x 10GbE, and a total of 6x 10GbEs are interconnected via 10G DAC cables to NSA 7141. The traffic is unidirectional with packet size of 64 bytes. In this test, each slot only uses two ports. One transmit queue is allocated for each port with affinity set to unique logical cores. Test results appear in Table VII. Compared with the theoretical maximum throughput frame size of 64 bytes (Table III), the TX rate reaches 100%.
Figure 6. Multiple NX 140F TX throughput test (2 ports per LAN module)
TABLE VII
TEST RESULTS FOR MULTIPLE NX 140F TX THROUGHPUT TEST (2 PORTS PER LAN MODULE)
Port |
Throughput (FPS) |
TX Rate (%) |
Slot1-Port 1 |
14880952 |
100% |
Slot1-Port 2 |
14880952 |
100% |
Slot2-Port 1 |
14880952 |
100% |
Slot2-Port 2 |
14880952 |
100% |
Slot3-Port 1 |
14880952 |
100% |
Slot3-Port 2 |
14880952 |
100% |
Testing results of the above configurations leads to the conclusion that NSA 7141 paired with NX 140F’s 10GbE ports can realize a 100% TX rate in 64-byte packet sizes, with utilization of up to two ports per LAN module and integrated with optimized DPDK, CPU core allocation, and software. The solution thus offers a stable 10GbE line rate as with other branded packet generators. Because of the collaboration between NEXCOM, a leading network and communication solution provider focused on high- performance and reliable computing platforms and network technology, and O’Prueba, dedicated to building easy-to-use and cost-effective testing solutions for the network and communications industry, state-of-the-art network performance benchmark testing solutions are now affordable, reliable, and serviceable.
References
[1] Spirent Communications, FS S5800-48F4S switch 1/10GE Ethernet performance validation. Calabasas: 2014. https://img-en.fs.com/file/report/fs-s5800-48f4s-1ge-and-10ge-ethernet-switch-performance-test-report.pdf
[2] Intel DPDK Validation Team, DPDK Intel NIC performance report release 18.02.
2018. https://fast.dpdk.org/doc/perf/DPDK_18_02_Intel_NIC_performance_report.pdf