3 $ ${bindir:=.}/ptask_L07_usage --cfg=host/model:ptask_L07 ${platfdir}/ptask_L07.xml --log=root.fmt=%m%n
4 > Configuration change: Set 'host/model' to 'ptask_L07'
5 > Switching to the L07 model to handle parallel tasks.
6 > TEST: Create and run a sequential execution.
7 > ------------------------------------------------------------
8 > Have to compute 1 flop on a 1 flop/s host.
9 > Should be done in exactly one second.
10 > Actual result: computing 1 flop at 1 flop/s takes 1.00 seconds.
13 > TEST: Create and run two concurrent sequential executions.
14 > ------------------------------------------------------------
15 > Have to compute 2 x 1 flop on a 1 flop/s host.
16 > Should be done in exactly 2 seconds because of sharing.
17 > Actual result: computing 2x1 flop at 1 flop/s takes 2.00 seconds.
20 > TEST: Create and run a parallel execution on 2 homogeneous hosts.
21 > ------------------------------------------------------------
22 > Have to compute 2 flops across two hosts running at 1 flop/s.
23 > Should be done in exactly one second.
24 > Actual result: computing 2 flops on 2 hosts at 1 flop/s takes 1.00 seconds.
27 > TEST: Create and run a parallel execution on 2 heterogeneous hosts.
28 > ------------------------------------------------------------
29 > Have to compute 2 flops across two hosts, one running at 1 flop/s and one at 2 flop/s.
30 > Should be done in exactly one second.
31 > Actual result: computing 2 flops on 2 heterogeneous hosts takes 1.00 seconds.
34 > TEST: Latency test between hosts connected by a shared link.
35 > ------------------------------------------------------------
36 > Have to send 1B from one host to another at 1Bps with a latency of 500ms.
37 > Should be done in 1.5 seconds (500ms latency + 1s transfert).
38 > Actual result: sending 1 byte on a shared link at 1Bps + 500ms takes 1.50 seconds.
41 > TEST: Latency test between hosts connected by a fatpipe link.
42 > ------------------------------------------------------------
43 > Have to send 1B from one host to another at 1Bps with a latency of 500ms.
44 > Should be done in 1.5 seconds (500ms latency + 1s transfert).
45 > Actual result: sending 1 byte on a fatpipe link at 1Bps + 500ms takes 1.50 seconds.
48 > TEST: Latency test between hosts connected by a 3-link route.
49 > ------------------------------------------------------------
50 > Have to send 1B from one host to another at 1Bps with a latency of 2 x 500ms + 1s.
51 > Should be done in 3 seconds (2 x 500ms + 1s latency + 1s transfert).
52 > Actual result: sending 1 byte on a 3-link route at 1Bps + 2,500ms takes 3.00 seconds.
55 > TEST: Latency test between hosts connected by a link with large latency.
56 > ------------------------------------------------------------
57 > Have to send 1B from one host to another on a link at 2Bps with a latency of 2 x 1024^2s.
58 > This latency is half the default TCP window size (4MiB). This limits the bandwidth to 1B
59 > Should be done in 2 x 1024^2s + 1 seconds (large latency + 1s transfert).
60 > Actual result: sending 1 byte on a large latency link takes 2097153.00 seconds.
63 > TEST: Latency test between hosts connected by a shared link with 2 comms in same direction.
64 > ------------------------------------------------------------
65 > Have to send 2 x 1B from one host to another at 1Bps with a latency of 500ms.
66 > Should be done in 2.5 seconds (500ms latency + 2s transfert).
67 > Actual result: sending 2x1 bytes on a shared link at 1Bps + 500ms takes 2.50 seconds.
70 > TEST: Latency test between hosts connected by a fatpipe link with 2 comms in same direction.
71 > ------------------------------------------------------------
72 > Have to send 2 x 1B from one host to another at 1Bps with a latency of 500ms.
73 > Should be done in 1.5 seconds (500ms latency + 1s transfert).
74 > Actual result: sending 2x1 bytes on a fatpipe link at 1Bps + 500ms takes 1.50 seconds.
77 > TEST: Latency test between hosts connected by a 3-link route with 2 comms in same direction.
78 > ------------------------------------------------------------
79 > Have to send 2 x 1B from one host to another at 1Bps with a latency of 2 x 500ms + 1s.
80 > Should be done in 4 seconds (2 x 500ms + 1s latency + 2s transfert).
81 > Actual result: sending 2x1 bytes on a 3-link route at 1Bps + 2,500ms takes 4.00 seconds.
84 > TEST: Latency test between hosts connected by a shared link with 2 comms in opposite direction.
85 > ------------------------------------------------------------
86 > Have to send 1B between two hosts in each direction at 1Bps with a latency of 500ms.
87 > Should be done in 2.5 seconds (500ms latency + 2s transfert).
88 > Actual result: sending 1 byte in both directions on a shared link at 1Bps + 500ms takes 2.50 seconds.
91 > TEST: Latency test between hosts connected by a fatpipe link with 2 comms in opposite direction.
92 > ------------------------------------------------------------
93 > Have to send 1B between two hosts in each direction at 1Bps with a latency of 500ms.
94 > Should be done in 1.5 seconds (500ms latency + 1s transfert).
95 > Actual result: sending 1 byte in both directions on a fatpipe link at 1Bps + 500ms takes 1.50 seconds.
98 > TEST: Latency test between hosts connected by a 3-link route with 2 comms in opposite direction.
99 > ------------------------------------------------------------
100 > Have to send 1B between two hosts in each direction at 1Bps with a latency of 2 x 500ms + 1s.
101 > Should be done in 4 seconds (2 x 500ms + 1s latency + 2s transfert).
102 > Actual result: sending 1 byte in both directions on a 3-link route at 1Bps + 2,500ms takes 4.00 seconds.
105 > TEST: 4-host parallel communication with independent transfers.
106 > ------------------------------------------------------------
107 > 'cpu0' sends 1B to 'cpu1' and 'cpu2' sends 1B to 'cpu3'. The only shared link is the fatpipe switch.
108 > Should be done in 3 seconds (2 x 500ms + 1s latency + 1s transfert).
109 > Actual result: sending 2 x 1 byte in a parallel communication without interference takes 3.00 seconds.
112 > TEST: 4-host parallel communication with scatter pattern.
113 > ------------------------------------------------------------
114 > 'cpu0' sends 1B to 'cpu1', 2B to 'cpu2' and 3B to 'cpu3'.
115 > Should be done in 8 seconds: 2 x 500ms + 1s of initial latency and :
116 > - For 3 seconds, three flows share a link to transfer 3 x 1B. 'cpu1' received its payload
117 > - For 2 seconds, two lows share a link to transfer 1 x 1B. 'cpu2' received is payload
118 > - For 1 second, one flow has the full bandwidth to transfer 1B. 'cpu3' received is payload
119 > Actual result: scattering an increasing number of bytes to 3 hosts takes 8.00 seconds.
122 > TEST: 4-host parallel communication with all-to-all pattern.
123 > ------------------------------------------------------------
124 > Each host sends 1B to every other hosts.
125 > Should be done in 8 seconds: 2 x 500ms + 1s of initial latency and 6 seconds for transfer
126 > Each SHARED link is traversed by 6 flows (3 in and 3 out).
127 > Each 1B transfer thus takes 6 seconds on a 1Bps link
128 > Actual result: 1-byte all-too-all in a parallel communication takes 8.00 seconds.
131 > TEST: Two concurrent communications, 1 large and 1 small.
132 > ------------------------------------------------------------
133 > A host sends two messages of 100MB and 1B to the other host.
134 > Should be done in 0.8001 seconds: 1e8/1.25e8 for transfer + 1e-4 of latency
135 > The small communication has a negligible impact on the large one.
136 > This corresponds to paying latency once and having the full bandwidth for the large communication.
137 > Actual result: 1 small and 1 large concurrent communications take 0.8001 seconds.
140 > TEST: Concurrent communication and computation.
141 > ------------------------------------------------------------
142 > A host sends 1B to another while the latter compute 2 flop.
143 > Should be done in 2 seconds: 1.5s to transfer and 2 seconds to compute.
144 > The two activities should overlap smoothly as they use different resources.
145 > The completion time is thus the maximum of the time to complete the two activities.
146 > Actual result: Sending 1B while computing 2 flops takes 2.0000 seconds.
