User manual MAXTOR 339MDWP1

[. . . ] 10 Test #8: Testing Two Cards. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13 Table of Contents I 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 List of Figures Figure 1: Figure 2: Figure 3: Figure 4: Figure 5: Figure 6: Throughput and Server CPU Utilization - Bench Configuration . . . . . . . . . . . . . . . 3 Throughput by Transfer and packet size - Bench Configuration. . . . . . . . . . . . . . . . 5 Adding IP Forwarding - Bench Configuration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Smartbits 1000 Test - Bench Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Combined Throughput Testing - Combined Throughput Setup. . . . . . . . . . . . . 10 Testing Two Cards - Bench Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 II List of Figures 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 List of Tables Table 1: Table 2: Table 3: Table 4: Table 5: Table 6: Table 7: Table 8: Table 9: Table 10: Server and Client Computer Descriptions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Throughput and Server CPU Utilization - Netperf Benchmark Results. . . . [. . . ] workstations. 2 Benchmark Utilities Description Performance Numbers Under Red Hat Linux 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 DETAILED BENCHMARKING TESTS AND RESULTS TEST #1: THROUGHPUT AND SERVER CPU UTILIZATION The first test used the NetPerf benchmarking utility to obtain the throughput as well as the server CPU utilization. A packet size of 8 kilobytes (8192 bytes) was used throughout the test. The test bench was set up as shown in Figure 1, below. Figure 1: THROUGHPUT AND SERVER CPU UTILIZATION - BENCH CONFIGURATION Linux PC with a Matrox NS-FNIC/4 Linux PC with a 10/ 100 Ethernet Card Subnet 1 Subnet 2 Subnet 3 Subnet 4 Linux PC with a 10/ 100 Ethernet Card Linux PC with a 10/ 100 Ethernet Card Linux PC with a 10/ 100 Ethernet Card For this test, the clients were all 166MHz Pentiums. The results for throughput and server CPU utilization are summarized in Table 2. The throughput is in megabits per second (1 million bits per second) and the CPU utilization is in percentage. TA B L E 2: THROUGHPUT AND SERVER CPU UTILIZATION - NETPERF BENCHMARK RESULTS Port 1 Port 2 Port 3 Port 4 (Throughput / (Throughput / (Throughput / (Throughput / CPU Utilization) CPU Utilization) CPU Utilization) CPU Utilization) 10Mbits/s Half duplex 10Mbits/s Full duplex 100Mbits/s Half duplex 100Mbits/s Full duplex 6. 86 / 1. 85 9. 38 / 2. 59 65. 98 / 17. 5 88. 81 / 23. 97 6. 87 / 1. 86 9. 38 / 2. 51 66. 14 / 17. 06 88. 37 / 22. 98 6. 89 / 1. 85 9. 38 / 2. 59 66. 02 / 17. 51 89. 43 / 24. 33 6. 89 / 1. 85 9. 38 / 2. 66 66. 03 / 17. 56 89. 51 / 24. 11 On average, a Matrox NS-FNIC/4 port delivers a throughput of 89Mbits/s with 24% of server CPU utilization at 100Mbits/s full duplex. Detailed Benchmarking Tests and Results 3 Performance Numbers Under Red Hat Linux 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 TEST #2: VALIDATION OF THROUGHPUT AND CPU UTILIZATION The second test used the same test bench as Test #1 (see Figure 1). The purpose of this test was to verify the throughput measurements of Test #1 with a second independent application. The only limitation of this test is that no CPU utilization measurements are provided. TTCP gives a throughput measurement by sending 16 megabytes (16, 777, 216 bytes) in packets of 8 kilobytes (8192 bytes) to the server. It then divides the total transfer size by the time it took and converts this result to Mbits/s. The throughput obtained, summarized in Table 3, is similar to that obtained from Test #1. The major difference between Test #1 and Test #2 occurs in the method in which the table results were computed. The results from Test #1 are averages computed by the utility itself after automatically performing the test many times. TTCP, however, only performs one data transfer and then displays the computed result. In order to get a reliable value, ten such tests were performed on each port and the arithmetic mean of those ten results was computed. Table 3 displays only the average results that were computed. TA B LE 3: VALIDATION OF THROUGHPUT - TTCP BENCHMARK RESULTS Port 1 (Mbits/s) 10Mbits/s Half duplex 10Mbits/s Full duplex 100Mbits/s Half duplex 100Mbits/s Full duplex 6. 49 8. 94 62. 34 89. 24 Port 2 (Mbits/s) 6. 51 8. 94 61. 56 89. 14 Port 3 (Mbits/s) 6. 54 8. 94 62. 29 89. 26 Port 4 (Mbits/s) 6. 53 8. 94 62. 05 88. 51 At 100Mbits/s full duplex, the results were identical at 89Mbits/s in both Test #1 (as shown in Table 2) and Test #2 (as shown in Table 3, above). At 100Mbits/s half duplex, the results were similar in both Test #1 (as shown in Table 2 at 66Mbits/s) and Test #2 (as shown in Table 3, above at 62Mbits/s). The difference was that the receiving part of the TTCP utility was not run on the server but on a client. In other words, IP forwarding was added to this test, as shown in Figure 3, below. Figure 3: ADDING IP FORWARDING - BENCH CONFIGURATION TCCP Application Client 1 Client 2 Client 3 Client 4 1 IP Forwarding 2 3 4 NS-FNIC/4 Ports in the Server This test verified whether the Matrox NS-FNIC/4 gives similar throughput when used in a router configuration. The reported throughput was similar to those reported by the first three tests and is shown in Table 5. This table only summarizes the results for a packet size of 8KB and a file size of 10MB. Table 5 displays the average of the results obtained in routing packets from one port to the other three. For example, the result displayed for port one is the average of the measured throughput on each of the three other ports. TA B L E 5: ADDING IP FORWARDING - TTCP ROUTING Port 1 (Mbits/s) 10Mbits/s Half duplex 10Mbits/s Full duplex 100Mbits/s Half duplex 100Mbits/s Full duplex 6. 32 8. 94 61. 14 61. 17 Port 2 (Mbits/s) 6. 32 8. 94 61. 13 61. 54 Port 3 (Mbits/s) 6. 28 8. 94 61. 32 63. 47 Port 4 (Mbits/s) 6. 28 8. 94 62. 00 64. 59 The results in Table 5 are similar to those of Table 4, showing that the Matrox NS-FNIC/4 gives similar performance results irrespective of the routing configuration. [. . . ] Since there are now two clients per port, both the bandwidth available and CPU Usage per client is now halved. · Detailed Benchmarking Tests and Results 11 Performance Numbers Under Red Hat Linux 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 TEST #8: TESTING TWO CARDS Test #8 used two Matrox NS-FNIC/4 cards in the server. As for Test #7, there were two 100Mhz Pentiums and two 166Mhz Pentiums on each Matrox NS-FNIC/4. Figure 6: TESTING TWO CARDS - BENCH SETUP Linux PC with a 10/100 Ethernet card Linux PC with a 10/100 Ethernet card Linux PC with a 10/100 Ethernet card Linux PC with a 10/100 Ethernet card subnet 1 card 1 subnet 1 Linux PC with a 10/100 Ethernet card Linux PC with a 10/100 Ethernet card Linux PC with a 10/100 Ethernet card Linux PC with a 10/100 Ethernet card subnet 2 subnet 2 Linux PC with 2 Matrox NS-FNIC/4 subnet 3 subnet 3 subnet 4 card 2 subnet 4 Since having eight ports active simultaneously is very demanding on the server, the expected results of this test were a little lower than the two previous tests. The combined throughputs were expected to be near the ones obtained for those same tests. The expected results were met, as shown in Table 10. TA B LE 10: TESTING TWO CARDS - COMBINED THROUGHPUT TEST Port 1 2 3 4 5 6 7 8 Combined throughput 100Mbits/s full duplex (Mbits/s) 34. 44 35. 12 34. 23 35. 00 34. 52 34. 67 34. 84 53. 98 296. 80 100Mbits/s half duplex (Mbits/s) 39. 88 39. 40 39. 82 39. 65 36. 42 39. 43 39. 73 46. 23 320. 56 The CPU utilization for this test can be described as follows: · · At 100Mbits/s full duplex, each port ranged from 9% to 15% averaging at 12. 38% and overall, it ranged from 72% to 100% averaging at 99%. [. . . ]