/gpu_cc_tb

Primary LanguageCuda

gCCTB

Overview

gCCTB is a testbed designed to evaluate various concurrency control schemes in GPU environments comprehensively. gCCTB allows for easy and efficient testing of different CC schemes on GPUs using a variety of benchmarks under different configurations.

Features

Supported CC Schemes

Scheme Name GPU Oriented Latch Version
2PL no-wait tpl_nw N slow_tpl_nw
2PL wait-die tpl_wd N slow_tpl_wd
TO to N slow_to
MVCC mvcc N slow_mvcc
Silo silo N slow_silo
TicToc tictoc N slow_tictoc
GaccO gacco Y
GPUTx gputx Y

Metrics

Metrics Unit Measured On
Throughput Tx/s CPU
Total Number of Rollbacks CPU
Average Total Duration s CPU
Average Precompute Duration s CPU
Average GPU Duration s CPU
Average GPU Duration ms GPU
Average Timestamp Allocation Duration ms GPU
Average Waiting Duration ms GPU
Average Aborting Duration ms GPU
Average Manager Duration ms GPU
Average Index Duration ms GPU

Requirements

  • NVIDIA GPU with compute capability 8.9
  • CUDA 12.0
  • Python3

Build

python3 compile.py

Useage

Dataset Generation

# format
./out/gen_ycsb {ycsb_rownum} {ycsb_txcnt} {theta} {r_ratio} {path_to_dataset}
# example
./out/gen_ycsb 10485760 1048576 0.1 0.9 ./dataset/ycsb.txs

where $theta \in [0,1.0], r_ratio \in [0,1.0]$

Benchmark

# format
./out/test_ycsb {path_to_dataset} {ycsb_rownum} {ycsb_txcnt} {cc_scheme} {warp_density} {block_size} {batch_size} {debug_flag}
# example
./out/test_ycsb ./dataset/ycsb.txs 10485760 1048576 tpl_nw 5 32 65536 0

where $warp_density \in [0,5], block_size \in [0,32]$

Output Explanation

{cc_scheme} {warp_density} {block_size} {batch_size} {throughput} {tot_abort_cnt} {avr_abort_cnt} {avr_ts_d} {avr_wait_d} {avr_abort_d} {avr_mng_d} {avr_tot_d} {avr_idx_d} {avr_pre} {avr_proc} {avr_tot}

where avr_tot_d corresponds to the Average GPU Duration measured on GPU, avr_proc corresponds to the Average GPU Duration measured on CPU, avr_tot corresponds to the Average Total Duration.

Experiment Data

Experiment data is listed in out_data directory.

TODO

  • TPC-C Support
  • More Indexes

Related Projects

DBx1000

Xiangyao Yu, George Bezerra, Andrew Pavlo, Srinivas Devadas, and Michael Stonebraker. 2014. Staring into the abyss: an evaluation of concurrency control with one thousand cores. Proc. VLDB Endow. 8, 3 (November 2014), 209–220. https://doi.org/10.14778/2735508.2735511

CCBench

Takayuki Tanabe, Takashi Hoshino, Hideyuki Kawashima, and Osamu Tatebe. 2020. An analysis of concurrency control protocols for in-memory databases with CCBench. Proc. VLDB Endow. 13, 13 (September 2020), 3531–3544. https://doi.org/10.14778/3424573.3424575