Reproducing the Experiment with CZNS and Balloon-ZNS

This repository provides instructions for reproducing the Experiment using FEMU with CCZNS and Balloon-ZNS

---

Setup Instructions

1. Prepare the VM Image by following the instructions in the FEMU repository

2. Clone the project folder from GitHub

3. To set up FEMU, navigate to the script folder of the respective method and run install_femu.sh

   If you want to test SSD without compression, use the Balloon-ZNS version of the software on the CCZNS version of FEMU

4. Go to the FEMU/build-femu folder and edit the device size in the run-zns.sh script

5. Start FEMU by running run-zns.sh

6. In FEMU emulated OS, clone the project folder from Github

7. Install the CCZNS version of the Linux kernel by running emulator_install_kernel.sh

   It is recommended to use the same kernel across all tests to ensure a fair comparison. Modifications made to the kernel to enable CCZNS, do not affect the performance

8. To install fio, run emulator_install_fio.sh in the respective folder

   fio tests can now be run from the fio folder using fio

   When testing CCZNS, use xnvme eninge, as it is the only engine adapted for CCZNS.

9. To install RocksDB, run emulator_install_rocksdb.sh

   To test RocksDB you can use db_bench located in the rocksdb folder, or ycsb located in the YCSB-cpp folder.

---

Detailed Instructions

FEMU

1. To set fixed_compression_ratio, add fixed_compression_ratio=x to the -device femu, argument. Example:

-device femu,devsz_mb=4096,femu_mode=3,fixed_compression_ratio=2

2. All scripts are tested on Ubuntu 20.04, while all scripts that are run inside FEMU are tested on Ubuntu 22.04, installed with instructions provided in the official FEMU repository: https://github.com/MoatLab/FEMU**{https://github.com/MoatLab/FEMU

fio tests

1. Set scheduler to deadline with:

sudo sh -c 'echo deadline > /sys/class/block/nvme0n1/queue/scheduler'

2. While performing fio tests, there are some arguments that are important:

   1. rw - write, read, and randread were used to perform sequential write, sequential read, and random read
   2. numjobs - All tests were performed with 16 jobs
   3. max_open_zones - In all tests a maximum of 14 open zones was allowed
   4. fixed_compression_ratio - Should be set to 1 when no compression is performed, otherwise, it should be equal the desired compression ratio
   5. md_per_io_size - In all tests this value was set to the minimum possible value

3. Reset all the zones before performing sequential write tests to ensure that no overhead from resetting zone will occurs

4. To perform sequential read or random read tests, the device must be pre-filled with data

5. An example command to perform sequential write test:

sudo ./fio --ioengine=xnvme --direct=1 --filename=/dev/nvme0n1 --rw=write --bs=64k --group_reporting --zonemode=zbd --name=seqwrite --numjobs=16 --offset_increment=8z --size=8z --max_open_zones=14 --md_per_io_size=1024 --thread=1  --fixed_compression_ratio=2

RocksDB tests

1. Set scheduler to deadline with:

sudo sh -c 'echo deadline > /sys/class/block/nvme0n1/queue/scheduler'

2. Create a rocksdb.properties and set following variables in it:
   1. rocksdb.fs_uri=zenfs://dev:nvme0n1
   2. rocksdb.dbname=/
   3. rocksdb.use_direct_io_for_flush_compaction=true
   4. rocksdb.use_direct_reads=true
   5. rocksdb.rocksdb.cache_size=4294967296
3. Create file-system with zenfs located in rocksdb/plugin/zenfs/util/zenfs:

sudo zenfs mkfs --zbd=nvme0n1 --aux_path=/tmp/zenfs-aux-1 --force

4. Put the dataset in a file dataset.txt in the same directory as ycsb executable, where each line contains a value for a key-value pair. Keys are generated automatically
5. An example command to perform workload A test:

sudo ./ycsb -load -run -db rocksdb -s -P workloads/workloada -P rocksdb.properties -threads 16