Artifact repository for Pavise

Pavise: Integrating Fault Tolerance Support for Persistent Memory Applications

Han Jie Qiu, Sihang Liu, Xinyang Song, Samira Khan, Gennady Pekhimenko 31th International Conference on Parallel Architectures and Compilation Techniques (PACT), 2022

We will provide access to our non-volatile memory server for artifact evaluation.

Figure 11 is our major result. To reproduce them:

git clone https://github.com/hjjq/pavise-pact22-artifact.git
cd pavise-pact22-artifact
# run experiments using screen to prevent progress loss.
# if for some reason the screen session is disconnected, use screen -r to reattach to the session
screen 
# repro.sh runs all key experiments
source repro.sh 
# "All experiments reproduced successfully" indicates completion

After repro.sh completes, the experiment results will be placed in $PAVISE_ROOT/results, where $PAVISE_ROOT is the root directory of the git repository.

$PAVISE_ROOT/results/summary.csv summarizes the results of all experiments. To view the results of each individual experiment, please see the rest of the files in $PAVISE_ROOT/results:

File in $PAVISE_ROOT/results	Corresponding results in Figure 11
microbench_ignorelist.csv	Pavise-ignore-list bars for hm-tx, ctree, btree, rbtree, rtree, and hm-atomic
microbench_conservative.csv	Pavise-conservative bars for hm-tx, ctree, btree, rbtree, rtree, and hm-atomic
microbench_no_pavise.csv	Original Application bars for hm-tx, ctree, btree, rbtree, rtree, and hm-atomic
{memcached-L|memcached-W|redis|vacation}_ignorelist	Pavise-ignore-list bars for {memcached-L | memcached-W | redis | vacation}
{memcached-L|memcached-W|redis|vacation}_conservative	Pavise-conservative bars for {memcached-L | memcached-W | redis | vacation}
{memcached-L|memcached-W|redis|vacation}_no_pavise	Original Application bars for {memcached-L | memcached-W | redis | vacation}

Summary CSV Format

$PAVISE_ROOT/results/summary.csv will have the format as shown below. Note that in Figure 11, all bars are plotted as relative runtime to the Original Appplication. For applications that measure throughput (redis, memcached-L, memcached-W), the relative runtime is computed using the inverse of the measured throughput. E.g. relative runtime of Pavise-ignore-list redis = (Original Appplication throughput) / (Pavise-ignore-list throughput)

Runtime applications (lower is better)

benchmark	Pavise-ignore-list runtime(s)	Pavise-conservative	Original Appplication
hashmap_atomic	19.921474	25.126959	10.527226
hashmap_tx	8.573652	10.705328	6.090657
ctree	8.632038	11.064522	5.503845
btree	7.677068	8.916856	3.266477
rbtree	17.290886	19.553796	7.733091
rtree	25.154142	27.932917	6.111267
vacation	5135	6179	1928

Throughput applications (higher is better)

benchmark	Pavise-ignore-list throughput	Pavise-conservative	Original Appplication
redis	15856.88	11927.03	29528.63
memcached-L	15848.01	15844.94	25623.63
memcached-W	12.3	9.3	22.0