Closed traversaro closed 3 years ago
Nicogene
commented
3 years ago Today I tried on my laptop the telemetryDeviceDumper after #118 then without the compression, attaching to the gazebo icub simulator running demoYoga.
Here is the data:
test_telemetry_preV0.1.0pt1.zip
test_telemetry_preV0.1.0pt2.zip
test_telemetry_preV0.1.0pt3.zip
It is divided for the limitations in the file size of GH.
This will be our term of comparison when we will enable the compression
Nicogene
commented
3 years ago Here https://github.com/robotology/yarp-telemetry/issues/116#issuecomment-822292398 and here https://github.com/robotology/yarp-telemetry/issues/116#issuecomment-822295725 we have the performances of saving buffers of integers without compression
Nicogene
commented
3 years ago This is the impact of the compression in the benchmarking (always in Release)
see #123
-------------------------------------------------------------------------------
Buffer Manager Test
Benchmarching section scalar int
-------------------------------------------------------------------------------
/home/ngenesio/robotology/yarp-telemetry/test/BufferManagerTest.cpp:261
...............................................................................
benchmark name samples iterations estimated
mean low mean high mean
std dev low std dev high std dev
-------------------------------------------------------------------------------
BufferOfInt-1000Samples-
oneVariable-1x1 100 24 2.196 ms
23.289 us 22.0523 us 29.0057 us
11.5012 us 1.30895 us 27.2576 us
BufferOfInt-10000Samples-
oneVariable-1x1 100 25 2.205 ms
189.423 us 188.531 us 190.38 us
4.71909 us 4.00428 us 5.67096 us
BufferOfInt-100000Samples-
oneVariable-1x1 100 14 2.2904 ms
3.84534 ms 3.82939 ms 3.88533 ms
121.131 us 59.8298 us 254.113 us
-------------------------------------------------------------------------------
Buffer Manager Test
Benchmarching section vector int
-------------------------------------------------------------------------------
/home/ngenesio/robotology/yarp-telemetry/test/BufferManagerTest.cpp:303
...............................................................................
benchmark name samples iterations estimated
mean low mean high mean
std dev low std dev high std dev
-------------------------------------------------------------------------------
BufferOfInt-1000Samples-
oneVariable-3x1 100 26 2.184 ms
23.7464 us 23.5317 us 24.0815 us
1.34468 us 939.276 ns 1.90603 us
BufferOfInt-10000Samples-
oneVariable-3x1 100 25 2.1775 ms
210.662 us 209.416 us 212.311 us
7.22334 us 5.66413 us 11.2294 us
BufferOfInt-100000Samples-
oneVariable-3x1 100 14 2.3044 ms
3.78658 ms 3.77177 ms 3.80836 ms
90.1273 us 66.2636 us 151.99 us
-------------------------------------------------------------------------------
Buffer Manager Test
Benchmarching section matrix int
-------------------------------------------------------------------------------
/home/ngenesio/robotology/yarp-telemetry/test/BufferManagerTest.cpp:350
...............................................................................
benchmark name samples iterations estimated
mean low mean high mean
std dev low std dev high std dev
-------------------------------------------------------------------------------
BufferOfInt-1000Samples-
oneVariable-3x2 100 25 2.1725 ms
27.0848 us 26.6739 us 27.8695 us
2.792 us 1.71097 us 4.27228 us
BufferOfInt-10000Samples-
oneVariable-3x2 100 25 2.21 ms
220.268 us 218.948 us 221.918 us
7.52388 us 5.84708 us 10.4822 us
BufferOfInt-100000Samples-
oneVariable-3x2 100 13 2.184 ms
4.25831 ms 4.2465 ms 4.26993 ms
59.8258 us 54.1948 us 66.3546 us-------------------------------------------------------------------------------
Buffer Manager Test
Benchmarching section scalar int
-------------------------------------------------------------------------------
/home/ngenesio/robotology/yarp-telemetry/test/BufferManagerTest.cpp:261
...............................................................................
benchmark name samples iterations estimated
mean low mean high mean
std dev low std dev high std dev
-------------------------------------------------------------------------------
BufferOfInt-1000Samples-
oneVariable-1x1 100 26 2.275 ms
7.06808 us 6.04746 us 9.04823 us
6.98648 us 4.12175 us 11.0471 us
BufferOfInt-10000Samples-
oneVariable-1x1 100 27 2.3112 ms
16.0549 us 15.14 us 19.5875 us
8.14875 us 1.68754 us 19.0175 us
BufferOfInt-100000Samples-
oneVariable-1x1 100 26 2.2828 ms
127.115 us 126.024 us 128.614 us
6.51239 us 5.06768 us 9.06838 us
-------------------------------------------------------------------------------
Buffer Manager Test
Benchmarching section vector int
-------------------------------------------------------------------------------
/home/ngenesio/robotology/yarp-telemetry/test/BufferManagerTest.cpp:310
...............................................................................
benchmark name samples iterations estimated
mean low mean high mean
std dev low std dev high std dev
-------------------------------------------------------------------------------
BufferOfInt-1000Samples-
oneVariable-3x1 100 26 2.3244 ms
8.48223 us 7.87719 us 9.18527 us
3.33351 us 2.91963 us 3.8745 us
BufferOfInt-10000Samples-
oneVariable-3x1 100 27 2.2896 ms
17.3417 us 17.2061 us 17.5309 us
811.123 ns 628.317 ns 1.08676 us
BufferOfInt-100000Samples-
oneVariable-3x1 100 26 2.2958 ms
168.912 us 165.697 us 174.555 us
21.257 us 14.0929 us 33.3633 us
-------------------------------------------------------------------------------
Buffer Manager Test
Benchmarching section matrix int
-------------------------------------------------------------------------------
/home/ngenesio/robotology/yarp-telemetry/test/BufferManagerTest.cpp:364
...............................................................................
benchmark name samples iterations estimated
mean low mean high mean
std dev low std dev high std dev
-------------------------------------------------------------------------------
BufferOfInt-1000Samples-
oneVariable-3x2 100 22 2.2704 ms
7.54132 us 7.31777 us 8.07836 us
1.66523 us 807.941 ns 3.28624 us
BufferOfInt-10000Samples-
oneVariable-3x2 100 26 2.2932 ms
22.2203 us 21.9476 us 22.6629 us
1.74048 us 1.17876 us 2.40411 us
BufferOfInt-100000Samples-
oneVariable-3x2 100 25 2.285 ms
264.646 us 260.733 us 268.992 us
20.9638 us 18.0406 us 26.2273 usI tried the same experiment of https://github.com/robotology/yarp-telemetry/issues/115#issuecomment-812452312 after enabling the compression here is the data:
test_telemetry_compressed_pt1.zip test_telemetry_compressed_pt2.zip test_telemetry_compressed_pt3.zip
Here is the data, it is ok there are no artifacts:
For 2 minutes of acquisition of encoders/accelerations/velocities of 32 joints we have:
~ 9.2 MB uncompressed ~ 5.7 MB compressed
The differences in computation seem more significant, we have to decide if we want to lose something in performances for saving some space.
cc @AlexAntn @traversaro @S-Dafarra @drdanz
S-Dafarra
commented
3 years ago Nice! Would it be possible to save in another thread without blocking the circular buffer?
drdanz
commented
3 years ago Since the impact is rather significant, I believe that it should be a user choice, we could have an option to select is compression should be enabled or not...
As suggested by @S-Dafarra .