| Age | Commit message (Collapse) | Author |
|---|
|
IOshark support to input filesystem ftracepooints (readpage(s))
to capture and replay mmap'ed IO (which aren't captured by strace).
The new script collect-straces-ftraces.sh enables the necessary
tracepoints, and collects ftraces and straces compiles these into
intermediate bytecodes which ioshark_bench can then execute.
Test: Generate several monkey based workloads, and use the script
mentioned above to generate bytecodes and feed those into
ioshark_bench.
Change-Id: Iae57f5fdcdce0fedfd85c8d64ee006cc8b5b470d
Signed-off-by: Mohan Srinivasan <srmohan@google.com>
|
|
1) Add support to IOshark to not pre-create files that the workload
read from /system, /vendor and other read-only partitions. Instead
we just read the underlying files directly.
2) This support is implemented under a new -q (quick) mode.
3) Other misc fixes to do a setrlimit(NO_FD) so we don't run out
of fds during test.
4) Also commiting raw strace files from several monkey runs so we
can compile those if we make on disk IOshark format changes in the
future.
Test: Run IOshark with different workloads to make sure that this
skipping of pre-creation works fine.
Change-Id: Ic025cb0162eca479d9f529846e619dd4b71aa490
Signed-off-by: Mohan Srinivasan <srmohan@google.com>
|
|
Add a summary (-s) option that dumps out the metrics from the run
on one single line for easier parseability. Requested by Shankhoneer
for easier integration of ioshark into tradefed.
Test: Test the new -s option (and combination with other options)
to make sure everything continues to work.
Change-Id: Ic27306a89bb501ce5b31f8df2ef9c583f56e9e4a
Signed-off-by: Mohan Srinivasan <srmohan@google.com>
|
|
IOshark uses /proc/diskstats to get various stats. It needs a
block device name for doing this. The block device was hardcoded
as sda (Marlin/Sailfish). Query the device productname and use
this to set the block device name instead. Needed to make IOshark
work on non-M/S device.
Test: Testing the change on marlin as well as bullhead (which
covers Sailfish, Shamu and Angler as well).
Change-Id: Id9d7071019112cede1fd5e5c1f2c62f2fc4b6e65
Signed-off-by: Mohan Srinivasan <srmohan@google.com>
|
|
Android.mk flags to build and install ioshark only on userdebug
builds, where it is installed in xbin/.
Test: Build user and userdebug, and make sure that (a) ioshark
only gets built on userdebug builds and installed under xbin/.
Change-Id: Ia2a367d9e33884cc46f65761b283d8b1eb9f280a
Signed-off-by: Mohan Srinivasan <srmohan@google.com>
|
|
ioshark is a repeatable application workload based storage benchmark.
ioshark takes as input strace files, compiles them into bytecodes that
you can then feed into the benchmark. This allows us to collect straces
for a chosen application workload (by stracing zygote/zygote64), and use
these to do A-B comparisons of storage features (anything including the
filesystem and below, including new hardware).
Test: Extra tests for ioshark aren't needed, ioshark reports detailed
stats at the end of the run, which tells us whether the ioshark run
was successful or not. In addition, we can strace ioshark_bench as it
runs to make sure it is launching IOs as we would expect it to.
Change-Id: I51fa9879dc11b526385eb7827ca2b4186662a463
Signed-off-by: Mohan Srinivasan <srmohan@google.com>
|
|
Benchmark." (Broke the MacOS build - will commit after I test that this builds for MacOS).
This reverts commit a5dc6d6fcab8dfe719c437e8a9111ba962df1c15.
Change-Id: I53bc71eeb1bc3ad7706baae589cef8a80ef94c35
|
|
ioshark is a repeatable application storage workload benchmark. ioshark
takes as input strace files, compiles them into bytecodes that you can
then feed into the benchmark. This allows us to collect straces for a
chosen application workload (by stracing zygote/zygote64), and use
these to do A-B comparisons of storage features (anything including the
filesystem and below, including new hardware).
Test: Extra tests for ioshark aren't needed, ioshark reports detailed
stats at the end of the run, which tells us whether the ioshark run
was successful or not. In addition, we can strace ioshark_bench as it
runs to make sure it is launching IOs as we would expect it to.
Change-Id: I70674ff8daac229b9f8e5d37780f15c7a31a7ca0
Signed-off-by: Mohan Srinivasan <srmohan@google.com>
|
|
This reverts commit 99c8f98f2a9c76fabb19cdbc5b495978cfc479b3.
Change-Id: Id205f175802d55b8f12342f82199db5743e479dc
|
|
Change-Id: Iefd18cc8ebfe462231292e30397467410e9f0511
|
|
Change-Id: I83f0a8b93e1e75b21f6e1de365da411313e71923
|
|
Change-Id: Ief46c02a29a4787191004a5cc2e616175a0384da
|
|
ioshark is a repeatable application storage workload benchmark. ioshark
takes as input strace files, compiles them into bytecodes that you can
then feed into the benchmark. This allows us to collect straces for a
chosen application workload (by stracing zygote/zygote64), and use
these to do A-B comparisons of storage features (anything including the
filesystem and below, including new hardware).
Test: Extra tests for ioshark aren't needed, ioshark reports detailed
stats at the end of the run, which tells us whether the ioshark run
was successful or not. In addition, we can strace ioshark_bench as it
runs to make sure it is launching IOs as we would expect it to.
Change-Id: Ieb0ecf5e32695ea95b361710aa31613f6ce63b97
Signed-off-by: Mohan Srinivasan <srmohan@google.com>
|
