Latency Results from Pallas MPI Benchmarks
IA-32 Measurements
We used 2 Dell PowerEdge 1750 servers with 2 hyperthreaded Intel Xeon 3.06 GHz processors (3 GB RAM, 512k L2 cache) running Linux 2.4.20-30smp kernels for all bandwidth measurements with PMB.
Each machine had InfiniHost MT23108 HCA (a1 silicon) on 64 bit PCIX 133MHz bus with thca-3.1 for Infiniband and LANai 10 on 64 bit PCIX 100MHz bus with gm-2.0.11 driver for Myrinet.
|
Msg Size (Bytes) |
Myrinet Latency (usec) |
TCP Latency (usec) |
Infiniband Latency (usec) |
|
|
|
|
|
|
|
|
0 |
8.1 |
60.05 |
5.37 |
|
|
1 |
8.29 |
58 |
5.42 |
|
|
2 |
8.17 |
57.94 |
5.38 |
|
|
4 |
8.28 |
57.98 |
5.55 |
|
|
8 |
7.93 |
58.17 |
5.33 |
|
|
16 |
8.08 |
58.33 |
5.41 |
|
|
32 |
8.36 |
58.56 |
6.35 |
|
|
64 |
8.63 |
61.83 |
6.51 |
|
|
128 |
9.04 |
60.5 |
6.78 |
|
|
256 |
9.94 |
64.68 |
7.32 |
|
|
512 |
11.56 |
72.96 |
8.3 |
|
|
1024 |
15.31 |
86.89 |
10.36 |
|
|
2048 |
22.48 |
112.87 |
12.22 |
|
|
4096 |
32.51 |
165.2 |
16.21 |
|
|
8192 |
53.15 |
281.27 |
24.17 |
|
|
16384 |
100.19 |
477.95 |
41.6 |
|
|
32768 |
166.3 |
774.78 |
60.23 |
|
|
65536 |
299.07 |
868.1 |
97.88 |
|
|
131072 |
563.76 |
1404.57 |
173.45 |
|
|
262144 |
1093.74 |
2504.61 |
415.18 |
|
|
524288 |
2154.96 |
4631.62 |
627.66 |
|
|
1048576 |
4274.14 |
9016.31 |
1238.97 |
|
|
2097152 |
8519.93 |
17410.48 |
2473.25 |
|
|
4194304 |
16996 |
34434.4 |
5496.45 |
|
|
|
|
Msg Size (Bytes) |
Myrinet Latency (usec) |
TCP Latency (usec) |
Infiniband Latency (usec) |
|
|
|
|
|
|
|
|
0 |
10.53 |
76.69 |
5.79 |
|
|
1 |
11.04 |
74.56 |
5.8 |
|
|
2 |
10.96 |
73.45 |
5.78 |
|
|
4 |
11.13 |
72.8 |
5.9 |
|
|
8 |
11.05 |
74.51 |
5.8 |
|
|
16 |
10.99 |
74.75 |
5.94 |
|
|
32 |
11.1 |
75.2 |
6.75 |
|
|
64 |
11.2 |
76.73 |
6.89 |
|
|
128 |
11.82 |
79.22 |
6.97 |
|
|
256 |
12.28 |
83.28 |
7.53 |
|
|
512 |
13.98 |
87.75 |
8.53 |
|
|
1024 |
17.23 |
103.34 |
10.6 |
|
|
2048 |
25.49 |
122.59 |
12.51 |
|
|
4096 |
38.19 |
173.73 |
18.16 |
|
|
8192 |
58.84 |
286.01 |
31.61 |
|
|
16384 |
115.98 |
897.9 |
61.07 |
|
|
32768 |
186.68 |
1013.46 |
98.93 |
|
|
65536 |
316.94 |
1562.4 |
174.54 |
|
|
131072 |
584.04 |
2272.94 |
326.46 |
|
|
262144 |
1112.82 |
3925.84 |
629.54 |
|
|
524288 |
2176.82 |
7237.74 |
1238.2 |
|
|
1048576 |
4311.33 |
14727.53 |
2450.18 |
|
|
2097152 |
8566.4 |
30543.05 |
4919.55 |
|
|
4194304 |
17117.5 |
60269.5 |
9955.6 |
|
|
|
|
Msg Size (Bytes) |
Myrinet Latency (usec) |
TCP Latency (usec) |
Infiniband Latency (usec) |
|
|
|
|
|
|
|
|
0 |
10.53 |
76.97 |
5.75 |
|
|
1 |
10.61 |
84.89 |
5.7 |
|
|
2 |
10.65 |
73.03 |
5.74 |
|
|
4 |
10.68 |
72.52 |
5.74 |
|
|
8 |
10.65 |
72.49 |
5.87 |
|
|
16 |
10.71 |
72.79 |
5.92 |
|
|
32 |
10.87 |
73.29 |
6.67 |
|
|
64 |
10.84 |
74.76 |
6.83 |
|
|
128 |
11.56 |
76.98 |
7 |
|
|
256 |
12.11 |
80.98 |
7.59 |
|
|
512 |
13.71 |
87.03 |
8.59 |
|
|
1024 |
17.13 |
102.36 |
10.56 |
|
|
2048 |
25.2 |
122.11 |
12.64 |
|
|
4096 |
37.99 |
186.31 |
17.78 |
|
|
8192 |
58.59 |
284.18 |
52.29 |
|
|
16384 |
116.52 |
1038.93 |
61.5 |
|
|
32768 |
182.03 |
1019.26 |
99.15 |
|
|
65536 |
315.35 |
1573.22 |
174.61 |
|
|
131072 |
600.61 |
2652.98 |
328.63 |
|
|
262144 |
1113.71 |
4818.84 |
628.82 |
|
|
524288 |
2174.84 |
9030.36 |
1235.07 |
|
|
1048576 |
4300.27 |
17614 |
2448.62 |
|
|
2097152 |
8985.6 |
34632.83 |
4907.53 |
|
|
4194304 |
17105.81 |
67963.41 |
9940.05 |
|
|
|
|
Msg Size (Bytes) |
Myrinet Latency (usec) |
TCP Latency (usec) |
Infiniband Latency (usec) |
|
|
|
|
|
|
|
|
0 |
18.73 |
95.87 |
11.24 |
|
|
1 |
18.84 |
96.21 |
11.5 |
|
|
2 |
18.83 |
96.34 |
11.41 |
|
|
4 |
18.78 |
96.27 |
11.38 |
|
|
8 |
18.81 |
96.25 |
11.33 |
|
|
16 |
18.81 |
96.47 |
11.39 |
|
|
32 |
18.89 |
97.06 |
12.9 |
|
|
64 |
19.05 |
98.9 |
12.89 |
|
|
128 |
19.18 |
100.4 |
13.22 |
|
|
256 |
19.79 |
104.45 |
13.88 |
|
|
512 |
20.97 |
112.9 |
15.1 |
|
|
1024 |
24.65 |
136.73 |
18.28 |
|
|
2048 |
35.29 |
138.55 |
21.46 |
|
|
4096 |
54.33 |
217.48 |
34.35 |
|
|
8192 |
97.08 |
391.92 |
56.11 |
|
|
16384 |
226.68 |
1015.34 |
120.21 |
|
|
32768 |
358.78 |
1421.01 |
195.86 |
|
|
65536 |
624.59 |
2623.98 |
346.87 |
|
|
131072 |
1157.33 |
4505.25 |
650.39 |
|
|
262144 |
2220.68 |
8528.48 |
1255.18 |
|
|
524288 |
4343.19 |
16620.89 |
2468.82 |
|
|
1048576 |
8662.19 |
32379.72 |
4897.59 |
|
|
2097152 |
17240.38 |
62717.04 |
9782.4 |
|
|
4194304 |
34221.1 |
126071.56 |
19866.44 |
|
|
|
IA-64 Measurements
To meaure bandwidth between IA-64 machines over Myrinet, we used 2 Intel Itanium 2 duals with 1.3 Ghz processors (4 GB RAM) running Suse SLES8 system with 2.4.19 kernel. GigE measurement were done with 2 Intel Itanium 2 duals with 900 Mhz processors (2 GB RAM) running 2.4.21 kernel.
For Myrinet, we used LANai 10.0 on 64 bit PCIX 100MHz bus with gm-2.0.8 driver. For Gigabit Ethernet, we had Broadcom Gigabit Ethernet adapter with bcm5700 driver.
|
Msg Size (Bytes) |
Myrinet Latency (usec) |
TCP Latency (usec) |
Infiniband Latency (usec) |
Shmem Latency (usec) |
|
|
|
|
|
|
|
0 |
11.43 |
73.35 |
8.04 |
1.63 |
|
1 |
12.29 |
72.54 |
8.22 |
1.68 |
|
2 |
12.48 |
73.59 |
8.25 |
1.75 |
|
4 |
12.48 |
72.42 |
8.14 |
1.78 |
|
8 |
12.15 |
73.71 |
8.29 |
1.72 |
|
16 |
12.96 |
72.57 |
8.9 |
1.73 |
|
32 |
13.27 |
74.93 |
9.11 |
1.79 |
|
64 |
12.73 |
73.65 |
9.25 |
2.1 |
|
128 |
13.23 |
77.87 |
9.76 |
2.27 |
|
256 |
14.26 |
80 |
10.28 |
2.55 |
|
512 |
16.67 |
89.6 |
11.62 |
3.08 |
|
1024 |
20.14 |
103.04 |
13.51 |
3.69 |
|
2048 |
28.15 |
122.5 |
15.76 |
4.75 |
|
4096 |
39.4 |
147.67 |
19.96 |
6.64 |
|
8192 |
63.54 |
171.59 |
28.83 |
10.28 |
|
16384 |
119.89 |
405.27 |
54.78 |
20.29 |
|
32768 |
189.2 |
539.36 |
76.96 |
33.96 |
|
65536 |
332.14 |
849.93 |
121.14 |
61.58 |
|
131072 |
617.02 |
1530.84 |
210.28 |
116.11 |
|
262144 |
1181.63 |
2793.1 |
387.92 |
228.78 |
|
524288 |
2330.24 |
4882.39 |
743.69 |
724.71 |
|
1048576 |
4562.07 |
9352.01 |
1453.71 |
1711.19 |
|
2097152 |
9073.9 |
18279.95 |
2903.43 |
3772.4 |
|
4194304 |
18321.3 |
36126.51 |
5774.45 |
6947.45 |
|
|
|
Msg Size (Bytes) |
Myrinet Latency (usec) |
TCP Latency (usec) |
Infiniband Latency (usec) |
Shmem Latency (usec) |
|
|
|
|
|
|
|
0 |
14.17 |
114.23 |
8.57 |
2.07 |
|
1 |
14.84 |
112.33 |
8.83 |
2.17 |
|
2 |
13.97 |
99.84 |
8.95 |
2.19 |
|
4 |
14.88 |
95.2 |
8.81 |
2.29 |
|
8 |
16.12 |
97.39 |
9 |
2.22 |
|
16 |
14.12 |
95.62 |
9.56 |
2.25 |
|
32 |
16.06 |
98.18 |
9.86 |
2.3 |
|
64 |
15.42 |
94.47 |
9.8 |
2.84 |
|
128 |
16.76 |
101.91 |
10.38 |
2.97 |
|
256 |
18.55 |
104.17 |
10.82 |
3.32 |
|
512 |
20.81 |
113.22 |
12.57 |
3.85 |
|
1024 |
23.07 |
124.98 |
14.09 |
4.49 |
|
2048 |
31.91 |
155.59 |
17.77 |
5.73 |
|
4096 |
45.36 |
169.74 |
22.69 |
8.14 |
|
8192 |
70.12 |
185.27 |
33.64 |
12.86 |
|
16384 |
137.99 |
439.16 |
78.38 |
25.42 |
|
32768 |
217.35 |
601.62 |
119.78 |
43.66 |
|
65536 |
362.95 |
1563.15 |
201.82 |
80.26 |
|
131072 |
696.82 |
2714.58 |
367.26 |
152.74 |
|
262144 |
1307.09 |
4947.29 |
695.85 |
317.81 |
|
524288 |
2426.44 |
9367.86 |
1358.75 |
1379.69 |
|
1048576 |
4691.18 |
17769.85 |
2683.75 |
3053.48 |
|
2097152 |
9311.04 |
35701.85 |
5359.9 |
6393.35 |
|
4194304 |
19711.8 |
62048.01 |
10631.2 |
12945.32 |
|
|
|
Msg Size (Bytes) |
Myrinet Latency (usec) |
TCP Latency (usec) |
Infiniband Latency (usec) |
Shmem Latency (usec) |
|
|
|
|
|
|
|
0 |
14.61 |
97.07 |
8.99 |
2.46 |
|
1 |
15.38 |
95.65 |
9.14 |
2.47 |
|
2 |
13.92 |
97.56 |
9.05 |
2.52 |
|
4 |
16.75 |
95.72 |
9.1 |
2.6 |
|
8 |
14.86 |
97.53 |
9.05 |
2.5 |
|
16 |
13.96 |
95.68 |
9.93 |
2.52 |
|
32 |
14.99 |
98.64 |
9.91 |
2.71 |
|
64 |
15.73 |
97.58 |
10.13 |
3.19 |
|
128 |
15.89 |
100.84 |
10.56 |
3.45 |
|
256 |
15.99 |
104.9 |
11.26 |
3.76 |
|
512 |
19.26 |
113.36 |
12.6 |
4.55 |
|
1024 |
21.89 |
128.46 |
14.61 |
5.38 |
|
2048 |
31.74 |
150.85 |
17.2 |
6.77 |
|
4096 |
44.88 |
170.37 |
22.73 |
9.58 |
|
8192 |
72.62 |
180.56 |
33.52 |
14.52 |
|
16384 |
141.06 |
597.55 |
77.21 |
27.75 |
|
32768 |
213.23 |
888.08 |
118.25 |
47.37 |
|
65536 |
365.31 |
1557.59 |
200.83 |
87.47 |
|
131072 |
658.52 |
2687.27 |
365.51 |
166.97 |
|
262144 |
1237.03 |
4914.76 |
694.97 |
332.14 |
|
524288 |
2445.89 |
9421.34 |
1357.82 |
1267.91 |
|
1048576 |
4662.65 |
18213.49 |
2680.54 |
2985.53 |
|
2097152 |
9256.12 |
35647.72 |
5314.8 |
6259.65 |
|
4194304 |
18379.4 |
63960.35 |
10627.4 |
12609.05 |
|
|
|
Msg Size (Bytes) |
Myrinet Latency (usec) |
TCP Latency (usec) |
Infiniband Latency (usec) |
Shmem Latency (usec) |
|
|
|
|
|
|
|
0 |
23.35 |
141.06 |
15.49 |
3.98 |
|
1 |
24.38 |
139.48 |
15.53 |
4.17 |
|
2 |
24.74 |
141.35 |
15.56 |
4.29 |
|
4 |
24.44 |
141.33 |
15.57 |
4.36 |
|
8 |
28.07 |
139.6 |
15.64 |
4.23 |
|
16 |
26.69 |
141.89 |
17.11 |
4.31 |
|
32 |
25.03 |
140.35 |
17.21 |
4.51 |
|
64 |
25.47 |
143.67 |
17.49 |
5.13 |
|
128 |
25.1 |
144.93 |
17.98 |
5.65 |
|
256 |
27.65 |
150.37 |
18.74 |
6.06 |
|
512 |
27.66 |
158.27 |
20.08 |
6.98 |
|
1024 |
30.48 |
136.73 |
23.43 |
7.89 |
|
2048 |
44.01 |
168.73 |
27.8 |
10.49 |
|
4096 |
66.87 |
236.95 |
37.84 |
15.96 |
|
8192 |
116.09 |
281.9 |
59.46 |
27.57 |
|
16384 |
270.06 |
923.39 |
160.13 |
63.07 |
|
32768 |
437.77 |
1787.94 |
245.2 |
111.31 |
|
65536 |
726.42 |
3067.43 |
414.45 |
206.99 |
|
131072 |
1310.2 |
5324.52 |
744.3 |
395.26 |
|
262144 |
2617.53 |
9784.23 |
1409.75 |
1089.17 |
|
524288 |
4875.97 |
18707.89 |
2745.47 |
2774.09 |
|
1048576 |
9670.91 |
36524.75 |
5395.29 |
6017.49 |
|
2097152 |
19320.05 |
71885.38 |
10702.43 |
12203.07 |
|
4194304 |
37614.31 |
125865.45 |
21416.95 |
24526.85 |
|
|
Back to the VMI2 web page.