README
1 This directory contains tools to profile the sdcard performance.
2
3 There are 2 parts to the tool:
4 * A binary that runs on the device, exercises the sdcard and send
5 measurment data back to the host (sdcard_perf_test).
6 * A host python script to plot the data.
7
8 Additionally, there is script 'profile_sdcard.sh' that allows you
9 to check how the sdcard scale with the number of processes.
10
11 INSTALLATION
12 ============
13 Build, install and mount debugfs. In this directory with a properly
14 configured enviroment:
15
16 mm SDCARD_TESTS=1
17 adb remount
18 adb push $ANDROID_PRODUCT_OUT/system/bin/sdcard_perf_test /system/bin/sdcard_perf_test
19 adb shell mount -t debugfs none /sys/kernel/debug
20
21 If you want to graph the results you need gnuplot and numpy:
22
23 sudo apt-get install gnuplot python-numpy python-numeric
24
25 You need Gnuplot.py version 1.8 (not the one coming with ubuntu).
26 Download it from the Gnuplot.py web site. Extract to a temp
27 directory, chdir and run:
28
29 sudo python setup.py install
30
31
32 INVOCATION
33 ==========
34
35 Run a simple test:
36
37 adb shell sdcard_perf_test --test=write --size=1000 --chunk-size=100 --procnb=1 --iterations=100
38
39 This test will write 1000kbyte (1M) files using writes of 100kbytes (so 10 writes per file) using
40 only 1 process for 100 times (100 files will be written on the sdcard).
41 The test will not call sync to flush the writes.
42 At the end of the test, some stats for the 'open' and 'write' system calls are written.
43
44 If you want to plot the data, you need to use the --dump option and provide a file:
45
46 adb shell sdcard_perf_test --test=write --size=1000 --chunk-size=100 --procnb=1 --iterations=100 --dump >/tmp/data.txt
47
48 PLOTTING
49 ========
50
51 To plot the result using the iter number of the x axis:
52
53 plot_sdcard.py -i /tmp/data.txt
54
55 To plot the result using time for the x axis:
56
57 plot_sdcard.py -t /tmp/data.txt
58
59 To plot the result from the profiler:
60
61 profile_sdcard.sh
62 plot_sdcard.py -p
63
64