Intel's Storage Performance Development Kit (SPDK) provides a set of tools and libraries for writing high performance, scalable, user-mode storage applications. It achieves high performance by moving all of the necessary drivers into userspace and operating in a polled mode instead of relying on interrupts, which avoids kernel context switches and eliminates interrupt handling overhead.
This fork of SPDK v16.06 provides a modified version of the nvme perf
example application to generate load for a local Flash performance test. The original application is modified to report read and write percentile latencies. The load generator is also modified to be open-loop, so you can sweep throughput by specifying a target IOPS instead of queue depth. You can use this load generator to precondition an SSD and/or run performance tests.
If you are using ReFlex, a software-based system that provides remote access to Flash at the performance of local Flash, you may find this load generator useful for preconditioning your SSD and running performance tests with different read/write ratios and request sizes to derive the request cost model.
Follow the instructions below if you are using ReFlex and have completed setup instructions 1-5 in the Reflex README. Otherwise follow instructions for general users in the next section.
# running an SPDK app requires installing DPDK
# we can use source code in reflex/deps/dpdk directory
cd PATH_TO_REFLEX/reflex/deps/dpdk
make install T=x86_64-native-linuxapp-gcc DESTDIR=.
cd ../../..
# fetch SPDK modified app code and compile
git clone https://github.com/anakli/spdk_perf.git
cd spdk_perf
make DPDK_DIR=../reflex/deps/dpdk/x86_64-native-linuxapp-gcc
# SPDK hugepage setup
sudo mkdir -p /mnt/huge
sudo mount -t hugetlbfs nodev /mnt/huge
cd examples/nvme/perf
# modify run_perf.sh with your desired rd/wr ratio, req size, number of cores, target IOPS, etc.
# run SPDK application to precondition SSD and/or do performance testing for cost model
sudo ./precondition.sh precond.csv
sudo ./run_perf.sh results.csv
# teardown spdk app hugepage setup to get ready to run ReFlex
sudo umount /mnt/huge
cat /proc/meminfo | grep Huge
# if no free hugepages, delete lingering hugepage files on other mount points for hugetlbfs
# $ cd /dev/hugepages
# $ rm -f rtemap*
cd PATH_TO_REFLEX/reflex
Doxygen API documentation is available, as well as a Porting Guide for porting SPDK to different frameworks and operating systems.
Many examples are available in the examples
directory.
To build SPDK, some dependencies must be installed.
Fedora/CentOS:
sudo dnf install -y gcc libpciaccess-devel CUnit-devel libaio-devel
# Additional dependencies for NVMf:
sudo dnf install -y libibverbs-devel librdmacm-devel
Ubuntu/Debian:
sudo apt-get install -y gcc libpciaccess-dev make libcunit1-dev libaio-dev
# Additional dependencies for NVMf:
sudo apt-get install -y libibverbs-dev librdmacm-dev
FreeBSD:
- gcc
- libpciaccess
- gmake
- cunit
Additionally, DPDK is required.
1) cd /path/to/spdk
2) wget http://dpdk.org/browse/dpdk/snapshot/dpdk-16.04.tar.gz
3) tar xfz dpdk-16.04.tar.gz
Linux:
4) (cd dpdk-16.04 && make install T=x86_64-native-linuxapp-gcc DESTDIR=.)
FreeBSD:
4) (cd dpdk-16.04 && gmake install T=x86_64-native-bsdapp-clang DESTDIR=.)
Once the prerequisites are installed, run 'make' within the SPDK directory to build the SPDK libraries and examples.
make DPDK_DIR=/path/to/dpdk
If you followed the instructions above for building DPDK:
Linux:
make DPDK_DIR=./dpdk-16.04/x86_64-native-linuxapp-gcc
FreeBSD:
gmake DPDK_DIR=./dpdk-16.04/x86_64-native-bsdapp-clang
Before running an SPDK application, some hugepages must be allocated and any NVMe and I/OAT devices must be unbound from the native kernel drivers. SPDK includes a script to automate this process on both Linux and FreeBSD. This script should be run as root.
sudo scripts/setup.sh
Example code is located in the examples directory. The examples are compiled automatically as part of the build process. Simply call any of the examples with no arguments to see the help output. You'll likely need to run the examples as a privileged user (root) unless you've done additional configuration to grant your user permission to allocate huge pages and map devices through vfio.