- Fedora 33 server
- Two 20-core Intel(R) Xeon(R) Gold 5218R CPU @ 2.10GHz
- 6 x 16GB DRAM per socket
- 6 x 128GB Intel Optane DC Persistent Memory per socket
MEMTIS currently supports two system configurations
- DRAM + Intel DCPMM (used only single socket)
- local DRAM + remote DRAM (used two socket, CXL emulation mode)
See linux/
You have to enable CONFIG_HTMM when compiling the linux source.
make menuconfig
...
CONFIG_HTMM=y
...
There are nothing special libraries for MEMTIS itself.
(You just need to install libraries for Linux compilation.)
See memtis-userspace/
Please read memtis-userspace/README.md for detailed explanations
- Reconfigures a namespace with devdax mode
sudo ndctl create-namespace -f -e namespace0.0 --mode=devdax
...
- Reconfigures a dax device with system-ram mode (KMEM DAX)
sudo daxctl reconfigure-device dax0.0 --mode=system-ram
...
We used open-sourced benchmarks except SPECCPU2017.
We provided links to each benchmark source in memtis-userspace/bench_dir/README.md
It is necessary to create/update a simple script for each benchmark. If you want to execute XSBench, for instance, you have to create memtis-userspace/bench_cmds/XSBench.sh.
This is a sample.
# memtis-userspace/bench_cmds/XSBench.sh
BIN=/path/to/benchmark
BENCH_RUN="${BIN}/XSBench [Options]"
# Provide the DRAM size for each memory configuration setting.
# You must first check the resident set size of a benchmark.
if [[ "x${NVM_RATIO}" == "x1:16" ]]; then
BENCH_DRAM="3850MB"
elif [[ "x${NVM_RATIO}" == "x1:8" ]]; then
BENCH_DRAM="7200MB"
elif [[ "x${NVM_RATIO}" == "x1:2" ]]; then
BENCH_DRAM="21800MB"
fi
# required
export BENCH_RUN
export BENCH_DRAM
cd memtis-userspace/
# check running options
./scripts/run_bench.sh --help
# create an executable binary file
make
# run
sudo ./scripts/run_bench.sh -B ${BENCH} -R ${MEM_CONFIG} -V ${TEST_NAME}
## or use scripts
sudo ./run-fig5-6-10.sh
sudo ./run-fig7.sh
...
See memtis-userspace/README.md
174ca88
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
To be updated
- Taehyung Lee (Sungkyunkwan University, SKKU) [email protected], [email protected]
- Sumit Kumar Monga (Virginia Tech) [email protected]
- Changwoo Min (Virginia Tech) [email protected]
- Young Ik Eom (Sungkyunkwan University, SKKU) [email protected]