Tests at Cambridge Site - NetFPGA/NetFPGA-public GitHub Wiki
This page displays the performance results for reference NIC 10G and 1G at the Cambridge site.
Performance Results - 10G NIC
Intel-10G NIC to/from Intel-10G NIC
Test Setup
This control test measures the performance of the Intel-10G X520-DA2 network interface cards (pictured below) as installed in our test hosts.
- The test setup uses two Core-i7 Linux hosts, each with an identical Intel PCIe NIC;
- The NICs are connected directly via a HP Procurve 10GBe SFP+ Direct Attach cable (model J9281B) as shown below:
- Host A has the fixed IP address 192.168.0.1 and runs the netserver component of netperf;
- Host B has the fixed IP address 192.168.0.2 and runs the netserver component of netperf;
- For each test, the tcp_stream_script and udp_stream_script netperf test scripts (located in the tools/scripts/ directory) were used;
- Four tests were conducted in all, using TCP and UDP with- and without forcing CPU affinity (i.e. forcing netperf processes to reside on CPU 1);
- The duration (in seconds) was reduced to 10 (from 60), resulting in a total test time of approximately 70 minutes.
Traffic from Host A (Intel 10G NIC) to Host B (Intel 10G NIC)
Please see below.
Traffic from Host B (Intel 10G NIC) to Host A (Intel 10G NIC)
| Test | Result |
|---|---|
| tcp_stream_script 192.168.0.1 CPU | cam_intel10G_to_intel10G_tcp_default.txt |
| udp_stream_script 192.168.0.1 CPU | cam_intel10G_to_intel10G_udp_default.txt |
| tcp_stream_script 192.168.0.1 CPU -T1,1 | cam_intel10G_to_intel10G_tcp_affinity.txt |
| udp_stream_script 1.1.1.2 CPU -T1,1 | cam_intel10G_to_intel10G_udp_affinity.txt |
Intel-10G NIC to/from NetFPGA-10G (10G Reference NIC)
Test Setup
This test measures the performance of the NetFPGA-10G (10G Reference NIC) connected directly to a Intel-10G X520-DA2 NIC.
- The test setup uses two Core-i7 Linux hosts, where Host A is fitted with the NetFPGA-10G card, programmed with the 10G Reference NIC and Host B is fitted with a Intel PCIe NIC. Host A uses the NetFPGA-10G Reference NIC driver (v1.3.2). The NICs are connected directly via a HP Procurve 10GBe SFP+ Direct Attach cable (model J9281B) as shown below:
- Host A has the fixed IP address 192.168.0.1 and runs| the netserver component of netperf;
- Host B has the fixed IP address 192.168.0.2 and runs the netserver component of netperf;
- For each test, the tcp_stream_script and udp_stream_script netperf test scripts (located in the tools/scripts/ directory) were used;
- Four tests were conducted in all, using TCP and UDP with- and without forcing CPU affinity (i.e. forcing netperf processes to reside on CPU 1);
- The duration (in seconds) was reduced to 10 (from 60), resulting in a total test time of approximately 70 minutes.
Traffic from Host A (NetFPGA-10G 10G Reference NIC) to Host B (Intel 10G NIC)
| Test | Result |
|---|---|
| tcp_stream_script 192.168.0.2 CPU | cam_nf10_to_intel10G_tcp_default.txt |
| udp_stream_script 192.168.0.2 CPU | cam_nf10_to_intel10G_udp_default.txt |
| tcp_stream_script 192.168.0.2 CPU -T1,1 | cam_nf10_to_intel10G_tcp_affinity.txt |
| udp_stream_script 192.168.0.2 CPU -T1,1 | cam_nf10_to_intel10G_udp_affinity.txt |
Traffic from Host B (Intel 10G NIC) to Host A (NetFPGA-10G 10G Reference NIC)
| Test | Result |
|---|---|
| tcp_stream_script 192.168.0.1 CPU | cam_intel10G_to_nf10_tcp_default.txt |
| udp_stream_script 192.168.0.1 CPU | cam_intel10G_to_nf10_udp_default.txt |
| tcp_stream_script 192.168.0.1 CPU -T1,1 | cam_intel10G_to_nf10_tcp_affinity.txt |
| udp_stream_script 192.168.0.1 CPU -T1,1 | cam_intel10G_to_nf10_udp_affinity.txt |
Performance Results - 1G NIC
Intel PRO/1000 PT Dual Port NIC to/from Intel PRO/1000 PT Dual Port NIC
Test Setup
This control test measures the performance of the Intel PRO/1000 PT Dual Port NICs as installed in our test hosts.
- The test setup uses two Core-i7 Linux hosts, where both Hosts A and B are fitted with on-board Intel PRO/1000 PT Dual Port NICs. Both machines use the Intel NIC driver incorporated into the Fedora 14 Linux kernel. The NICs are connected directly via a Category 5e ethernet patch cable as shown below:
- Host A has the fixed IP address 192.168.0.1 and runs the netserver component of netperf;
- Host B has the fixed IP address 192.168.0.2 and runs the netserver component of netperf;
- For each test, the tcp_stream_script and udp_stream_script netperf test scripts (located in the tools/scripts/ directory) were used;
- Four tests were conducted in all, using TCP and UDP with- and without forcing CPU affinity (i.e. forcing netperf processes to reside on CPU 1);
- The duration (in seconds) was reduced to 10 (from 60), resulting in a total test time of approximately 70 minutes.
Traffic from Host A (Intel 1G NIC) to Host B (Intel 1G NIC)
Please see below.
Traffic from Host B (Intel 1G NIC) to Host A (Intel 1G NIC)
| Test | Result |
|---|---|
| tcp_stream_script 192.168.0.1 CPU | cam_intel1G_to_intelG_tcp_default.txt |
| udp_stream_script 192.168.0.1 CPU | cam_intel1G_to_intel1G_udp_default.txt |
| tcp_stream_script 192.168.0.1 CPU -T1,1 | cam_intel1G_to_intel1G_tcp_affinity.txt |
| udp_stream_script 192.168.0.1 CPU -T1,1 | cam_intel1G_to_intel1G_udp_affinity.txt |
Intel PRO/1000 PT Dual Port NIC to/from NetFPGA-10G (1G Reference NIC)
Test Setup
This test measures the performance of the NetFPGA-10G (1G Reference NIC) connected to Intel PRO/1000 PT Dual Port NIC installed in our test hosts.
- The test setup uses two Core-i7 Linux hosts, where Host A is fitted with the NetFPGA-10G card, programmed with the 1G Reference NIC and Host B has a built-in RealTek Intel PRO/1000 Dual Port NIC. Host A uses the NetFPGA-10G Reference NIC driver (v1.3.2). The NICs are connected directly via a Category 5e ethernet patch cable as shown below:
- Host A has the fixed IP address 192.168.0.1 and runs the netserver component of netperf;
- Host B has the fixed IP address 192.168.0.2 and runs the netserver component of netperf;
- For each test, the tcp_stream_script and udp_stream_script netperf test scripts (located in the tools/scripts/ directory) were used;
- Four tests were conducted in all, using TCP and UDP with- and without forcing CPU affinity (i.e. forcing netperf processes to reside on CPU 1);
- The duration (in seconds) was reduced to 10 (from 60), resulting in a total test time of approximately 70 minutes.
Traffic from Host A (NetFPGA-10G 1G Reference NIC) to Host B (Intel 1G NIC)
| Test | Result |
|---|---|
| tcp_stream_script 192.168.0.2 CPU | cam_nf10_to_intel1G_tcp_default.txt |
| udp_stream_script 192.168.0.2 CPU | cam_nf10_to_intel1G_udp_default.txt |
| tcp_stream_script 192.168.0.2 CPU -T1,1 | cam_nf10_to_intel1G_tcp_affinity.txt |
| udp_stream_script 192.168.0.2 CPU -T1,1 | cam_nf10_to_intel1G_udp_affinity.txt |
Traffic from Host B (Intel 1G NIC) to Host A (NetFPGA-10G 1G Reference NIC)
| Test | Result |
|---|---|
| tcp_stream_script 192.168.0.1 CPU | cam_intel1G_to_nf10_tcp_default.txt |
| udp_stream_script 192.168.0.1 CPU | cam_intel1G_to_nf10_udp_default.txt |
| tcp_stream_script 192.168.0.1 CPU -T1,1 | cam_intel1G_to_nf10_tcp_affinity.txt |
| udp_stream_script 192.168.0.1 CPU -T1,1 | cam_intel1G_to_nf10_udp_affinity.txt |
