paul614 / MLstorage

Storage based on machine learning

MLstorage (Machine Learning based Storage)

This project aims to support a solid state hybrid drive(SSHD) with an intelligent cache module, which controls storage policy based on estimated IO access pattern.

Build

• enter kmod directory and just make
  • The kernel module will be built against a current kernel version.
  • If you try to build with another version, please define KSRC environment variable
  • Linux source headers should be installed before build starts.
• If build succeeds, kmod/mlstor.ko will be generated.

Run

• A MLstorage control script is provided as script/mlstorctl.sh.
• Prepare two partitions or devices for HDD and NVMe
  • Assumes /dev/sdb4 and /dev/nvme0n1p1
  • root permission required
• create MLstorage device
  • Usage: mlstorctl -c -b <backing device> -C <cache device> <mlstor name>
  • eg) `script/mlstorctl -c -b /dev/sdb4 -C /dev/nvme01n1p1 mlstor_test
• create file system(ext4) on a MLstorage
  • mkfs.ext4 /dev/mapper/mlstor_test
  • mkdir /data.mlstor
  • mount /dev/mapper/mlstor /data/.mlstor
• Run benchmarks
  • We provide useful fio(Flexible I/O) scripts
  • cd fio
  • ./runfio.sh -t mlstor_test -o output_mlstor.txt
  • cat output_mlstor.txt

---

_{This project has been supported by ICT Research and Development Program of MSIP/IITP(Developing System Software Technologies for Emerging New Memory That Adaptively Learn Workload Characteristics) under Grant 2019-0-00074.}