Thank you for your time on evaluating our artifact. Here are the steps for setting up the FEMU for CVSSD. We target Available and Functional flag because full experiments will take serveral months.

Contact: Ziyang Jiao ([email protected]), Xiangqun Zhang ([email protected])

• 0. Hardware requirements
• 1. Connecting into lab machine
• 2. Downloading the repository
• 3. Compilation
• 4. Preparing the VM image (skip for AE)
• 5. Starting the virtual machine
• 6. Connecting into the virtual machine
• 7. Setting up CV-SSD
• 8. FIO experiments on CV-SSD
• 9. FIO experiments on CVSS (CV-FS+CV-SSD+CV-Manager)

Please make sure you have at least 160 GiB memory and 150 GiB free space on your disk if testing on your own machine. Our evaluation is based on the following hardware specifications:

To connect to our lab machine and perform evaluation, we first need to connect to our university network:

ssh [email protected] -p 10500

Then we can connect to our lab machine:

ssh [email protected]

The passwords are included in the submission page.

Please make sure to create your working directory first under /media/tmp_nvme4/fast24ae/ directory:

cd /media/tmp_nvme4/fast24ae/
mkdir reviewer1
cd reviewer1

To clone the code, please run the following command in your working directory:

git clone https://github.com/ZiyangJiao/FAST24_CVSS_FEMU.git

To compile the code, please run the following commands after cloning the repository:

cd FAST24_CVSS_FEMU
mkdir build-femu
cd build-femu
cp ../femu-scripts/femu-copy-scripts.sh ./
./femu-copy-scripts.sh ./
./femu-compile.sh

You can either build your own VM image or use the VM image provided by us.

Option 1: This is the recommended way to get CV-SSD running quickly - Use our VM image file. To obtain the VM image, you can contact Ziyang Jiao, Email: [email protected] or Xiangqun Zhang, Email: [email protected]. The VM image downloading instructions will be sent to your email address.

Option 2: To build your own VM image, please refer to the FEMU instructions.

We first copy the pre-configured disk image to the current directory (working_directory/FAST24_CVSS_FEMU/build-femu):

cp /media/tmp_nvme4/fast24ae/u20s.qcow2.FAST24AE ./u20s.qcow2.FAST24AE

This step could take around 10 mins. The pre-configured disk image has installed all necessary modules and dependencies for the experiments, including our kernel level modification (https://github.com/ZiyangJiao/FAST24_CVSS_Kernel) and CV-FS (https://github.com/ZiyangJiao/FAST24_CVSS_CVFS).

To start the virtual machine, please run:

./run-blackbox.sh

This will start the virtual machine (based on QEMU). You can set the path to your VM image via IMGDIR=/ in the script.

The username and password for the VM are femu

Note: if you encounter errors with port 8080, then 8080 is occupied by other user(s). Please modify the port number in line 32 of run-blackbox.sh file (i.e., -net user,hostfwd=tcp::8080-:22 ) to use another port instead (e.g., 8181, 8282, 8383, etc...) and then run ./run-blackbox.sh

Although the terminal shows an operable console for the virtual machine, it has some limitations. For example, there are no highlights for the terminal. Using Ctrl-C could terminate the virtual machine directly instead of terminating the process running under the virtual machine. Therefore, we have an extra SSH port to connect to the virtual machine for better usability.

To connect to the virtual machine, please run on another terminal:

ssh [email protected] -p 10500
ssh [email protected]
ssh femu@localhost -p 8080(or the modified port number)

We use fio as an example here to test the functionality of CV-SSD:

• First, we check the status of the emulated device (/dev/nvme0n1) by issuing lsblk command. We should find the emulated device with ~120GiB capacity as below.

  femu@fvm:~$ lsblk 
  NAME    MAJ:MIN RM   SIZE RO TYPE MOUNTPOINT
  fd0       2:0    1     4K  0 disk 
  loop0     7:0    0  55.7M  1 loop /snap/core18/2785
  loop1     7:1    0  63.5M  1 loop /snap/core20/2015
  loop2     7:2    0  40.9M  1 loop /snap/snapd/19993
  loop3     7:3    0  55.7M  1 loop /snap/core18/2790
  loop4     7:4    0  91.9M  1 loop /snap/lxd/24061
  loop6     7:6    0  91.8M  1 loop /snap/lxd/23991
  loop7     7:7    0  40.9M  1 loop /snap/snapd/20290
  loop8     7:8    0  63.9M  1 loop /snap/core20/2105
  sda       8:0    0    80G  0 disk 
  ├─sda1    8:1    0     1M  0 part 
  └─sda2    8:2    0    80G  0 part /
  nvme0n1 259:0    0 119.2G  0 disk 

• Second, we initiate the capacity-variant mode and disable WL in the device.

  sudo nvme admin-passthru /dev/nvme0n1 --opcode=0xef --cdw10=0x0008 --cdw11=0x00 --cdw12=0x02 --cdw13=0x00 -r -b

This new nvme-cli command enables the CV-SSD mode, where cdw12 adjusts the acceleration factor for aging, and cdw13 update the threshold to determine retired blocks. A lower value makes the blocks retire earlier and thus causes a shorter device lifetime and higher device reliability. The CV-SSD will use the default threshold and ignore this field if cdw13=0x00.

• To run FIO directly:

  sudo fio --randrepeat=1 --ioengine=libaio --direct=1 --name=test --bs=16k --iodepth=128 --readwrite=randwrite  --size=10G --filename=/dev/nvme0n1

This example issues 10GiB random write to the CV-SSD (/dev/nvme0n1). The detailed usage of FIO can be found on https://fio.readthedocs.io/en/latest/fio_doc.html.

• We can manually set and unset the degraded mode for CV-SSD. This mode updates the block management policies in CV-SSD.
• To set:

  sudo nvme admin-passthru /dev/nvme0n1 --opcode=0xef --cdw10=0x0A --cdw11=0x01 -r -b

• To unset:

  sudo nvme admin-passthru /dev/nvme0n1 --opcode=0xef --cdw10=0x0A --cdw11=0x00 -r -b

This part tests the funtionality of CV-SSD and our hardware emulation platform. The CV-SSD exports a fixed logical capacity without the capacity variance feature since CV-FS and CV-Manager was not enabled. To enable capacity variance, please refer to the following section.

We now test the functionality of CV-FS:

• First, the CV-FS code can be found under the directory /home/femu/linux-5.15.0-76-generic-f2fs.

• Second (skip for AE), we can complie CV-FS using the run.sh script under this directory. This step can be skipped since the provided image has done it.

  sudo ./run.sh

• Third, we build CV-FS on CV-SSD. This can be done by running:

  inscvfs
  diskcvfs

This command will build the file system and mount the CV-SSD on the path /mnt/nvme0n1. It also creates directories for the experiments.

• Fourth, we can check the system status using lsblk. We should find that the device has been mounted successfully and is ready to use. Below is an example:

  femu@fvm:/mnt/nvme0n1$ lsblk
  NAME    MAJ:MIN RM   SIZE RO TYPE MOUNTPOINT
  fd0       2:0    1     4K  0 disk 
  loop0     7:0    0  55.7M  1 loop /snap/core18/2785
  loop1     7:1    0  63.5M  1 loop /snap/core20/2015
  loop2     7:2    0  40.9M  1 loop /snap/snapd/19993
  loop3     7:3    0  55.7M  1 loop /snap/core18/2790
  loop4     7:4    0  91.9M  1 loop /snap/lxd/24061
  loop6     7:6    0  91.8M  1 loop /snap/lxd/23991
  loop7     7:7    0  40.9M  1 loop /snap/snapd/20290
  loop8     7:8    0  63.9M  1 loop /snap/core20/2105
  sda       8:0    0    80G  0 disk 
  ├─sda1    8:1    0     1M  0 part 
  └─sda2    8:2    0    80G  0 part /
  nvme0n1 259:0    0 119.2G  0 disk /mnt/nvme0n1

We now test the functionality of CVSS and CV-Manager:

• First, we can find the code of CV-Manager (cv_manager.py) under /home/femu/f2fs and the user-level tool under /home/femu/f2fs-tools.

• Second, we can issue some workloads to the system. Below is an example:

  sudo fio --randrepeat=1 --ioengine=libaio --direct=1 --name=test --bs=16k --iodepth=128 --readwrite=randwrite  --size=10G --filename=/mnt/nvme0n1/test1.fio

  Note that the filename attribute is the mount point of the device, instead of /dev/nvme0n1.

• Third, the logical capacity of CVSS can be adjusted online by issuing:

  sudo /home/femu/f2fs/reduction_with_parameter.out 118

  The parameter (e.g., 118) is the new logical capacity we set for the system. This can also be done by using the file system utility tool.

  sudo cvfs.f2fs /dev/nvme0n1 -f /mnt/nvme0n1/resize.tmp -t 118

• Fourth, we can validate the size of the logical capacity by issuing:

  df -h /dev/nvme0n1

  The new logical size should be shown under the Size field similar as follows.

  Filesystem      Size  Used Avail Use% Mounted on
  /dev/nvme0n1    118G  1.9G  117G   2% /mnt/nvme0n1

• Fifth, we test the functionality of CVSS after the logical capacity is updated.

  sudo fio --randrepeat=1 --ioengine=libaio --direct=1 --name=test --bs=16k --iodepth=128 --readwrite=randwrite  --size=10G --filename=/mnt/nvme0n1/test2.fio

  We now should find the file (test2.fio) under the directory /mnt/nvme0n1/.

• Sixth, the monitor of bad capacity and the adjustment of the logical capacity can be automated by CV-Manager:

  sudo /home/femu/f2fs/cv_manager.py