The previous implementation tried to maintain a ring buffer of completed I/O retrieved from ->getevents() that had not yet been retrieved by ->event(), but this is not necessary. None of the other FIO engines implement this sort of ring buffer, and the only code that calls ->getevents() will always retrieve the full number of events via ->event() afterwards. Instead, just count how many I/Os have been completed within each ->getevents() call in iocq_count, and reset iocq_count on each new ->getevents(). Change-Id: I2776c3a5c89a0d2b24478328bc138ff7a9fb5bdd Signed-off-by: Daniel Verkamp <daniel.verkamp@intel.com> |
||
---|---|---|
.. | ||
.gitignore | ||
example_config.fio | ||
fio_plugin.c | ||
Makefile | ||
README.md |
Compiling
First, clone the fio source repository from http://github.com/axboe/fio
git clone http://github.com/axboe/fio
Then check out the fio 2.17 tag
cd fio && git checkout fio-2.17
Finally, compile the code with
./configure && make
Next, edit the CONFIG file located in the root of the SPDK repository and set CONFIG_FIO_PLUGIN to y and FIO_SOURCE_DIR to the location of the fio repository that was just created.
Further, you'll need to build DPDK with -fPIC set. You can do this by modifying your DPDK config file (i.e. config/defconfig_x86_64-native-linuxapp-gcc) to include the line
EXTRA_CFLAGS=-fPIC
At this point, build SPDK as per normal. The fio plugin will be placed in the same directory as this README.
Usage
To use the SPDK fio plugin with fio, simply set the following in the fio configuration file (see example_config.fio in the same directory as this README).
ioengine=<path to fio_plugin binary>
To select NVMe devices, you simply pass an identifier as the filename in the format
domain.bus.slot.func/namespace
Remember that NVMe namespaces start at 1, not 0! Also, the notation uses '.' throughout, not ':'. For example - 0000.04.00.0/1.
Currently the fio_plugin is limited to a single thread, so only one job is supported.
When testing random workloads, it is recommended to set norandommap=1. fio's random map processing consumes extra CPU cycles which will degrade performance over time with the fio_plugin since all I/O are submitted and completed on a single CPU core.