Skip to content

Latest commit

 

History

History
131 lines (97 loc) · 4.38 KB

README.md

File metadata and controls

131 lines (97 loc) · 4.38 KB
Raw

Perseid

This repository contains the artifact for the USENIX ATC'23 paper: "Revisiting Secondary Indexing in LSM-based Storage Systems with Persistent Memory".

For USENIX ATC'23 Artifact Evaluation, please check out the Comments for AEC from the atc23ae hotcrp system.

Repository contents

This prototype is built based on PebblesDB.

  • src/include/pebblesdb/db.h: DB interface (with secondary indexing)
  • src/db/: implementation of PebblesDB, Perseid, and LSM-based secondary index
  • src/db/sec_idx/: PM-based secondary indexes
  • src/db/sec_idx/[clht & CCEH]: the hybrid PM-DRAM validation
  • src/db/sec_idx_bench.cc: benchmark for secondary indexes
  • scripts: main evaluation scripts

Dependencies

  • libraries

    • Dependencies required by PebblesDB
    • PMDK (libpmem & libpmemobj)
    • google gflags

  • utilities for experiments:

    • numactl

System configuration for PM

1. set Optane DCPMM to AppDirect mode
$ sudo ipmctl create -f -goal persistentmemorytype=appdirect

2. configure PM device to fsdax mode
$ sudo ndctl create-namespace -m fsdax

3. create and mount a file system with DAX
$ sudo mkfs.ext4 -f /dev/pmem0
$ sudo mount -o dax /dev/pmem0 /mnt/pmem0

Building the benchmarks

We recommend that you use the scripts from the scripts directly. The scripts will build and run experiments automatically.

Quick start:

$ mkdir -p build && cd build
$ cmake -DCMAKE_BUILD_TYPE=Release .. 
$ make sec_idx_bench -j

The benchmark receives arguments with gflags, examples:

./sec_idx_bench --db=path_to_db --preload=path_to_preload --trace=path_to_trace

To facilitate evaluation, we use cmake to set most configurations for systems. Please check out the sec_idx_config.h.in and the CMakeLists.txt for more details.

Running experiments

scripts
|---- kick_the_tires.sh     # script of "Hello world"-sized examples
|---- eval_fig*.sh          # script for each figure in the paper
|---- sub/eval_fig*.sh      # scripts for subfigure (if you want to evaluate them separately)

For USENIX ATC'23 artifact evaluation kick-the-tires

Run kick_the_tires.sh and this script will build and run several systems. You should see Kick-the-tires passed! at the end if all procedures succeeded.

cd scripts
./kick_the_tires.sh

Usage

To run experiments for Perseid:

# cd scripts  # in the scripts directory
./eval_fig*.sh

If you want to evaluate a subfigure separately, use the scripts in sub directory but stay in the scripts directory:

./sub/eval_fig*.sh

If you want to evaluate other systems, you can use the following commands:

./eval_fig*.sh others   # (just type 'others') for other systems (LSMSI, composite indexes, etc)
# or
./sub/eval_fig*.sh others

Results are shown in screen and saved in results directory with name like fig*_sysname.txt. Results are printed per 1 Million ops by default, and final results are printed after each experiment.

Each line contains several numbers:

  • Number of operations (in Million)
  • avg latency of all operations
  • avg latency of PUT
  • avg latency of GET by primary key
  • avg latency of GET by secondary key
  • avg latency of lookup LSM primary table in one secondary GET
  • avg latency of validation
  • avg number of query results of one secondary GET
  • avg number of primary table components probed in one secondary GET

You can use grep_result.sh to grep final results of latency and throughput:

./grep_result.sh path_to_result

Note:

  • These scripts should be run inside the scripts directory.
  • eval_fig7.sh will also get results of fig6, so you can skip eval_fig6.sh to save time.

Estimated time:

  • fig6 (not recommend): 1 hour for Perseid (6 results) and 3 hour for others (18 results)
  • fig7: 1 hour for Perseid (12 results) and 3.8 hours for others (36 results)
  • fig8: 35 minutes for Perseid (3 results) and 1.7 hours for others (6 results)
  • fig9: 4.3 hours for Perseid (12 results) and 7 hours for others (24 results)
  • fig10: 15 minutes for Perseid (10 results) and 2 hours for others (20 results)
  • fig11: 33 minutes for Perseid (3 results) and 2.6 hours for others (6 results)