void VsStorageOverheadThreshold::process()
{
cerr << "This is an experiment with name: "
<< this->getClassName() << endl;
SimulationConf simConf;
SchemaStats stats;
Cost cost(stats);
util::AutoTimer timer;
ExperimentalData queryIOExp("QueryIOVsStorageOverheadThreshold");
ExperimentalData runningTimeExp("RunningTimeVsStorageOverheadThreshold");
ExperimentalData storageExp("StorageOverheadVsStorageOverheadThreshold");
auto expData = { &queryIOExp, &runningTimeExp, &storageExp };
makeQueryIOExp(&queryIOExp);
makeRunningTimeExp(&runningTimeExp);
makeStorageExp(&storageExp);
for (auto exp : expData)
exp->open();
auto solvers = {
SolverFactory::instance().makeSinglePartition(),
SolverFactory::instance().makePartitionPerAttribute(),
SolverFactory::instance().makeOptimalOverlapping(),
SolverFactory::instance().makeOptimalNonOverlapping(),
SolverFactory::instance().makeHeuristicOverlapping(),
SolverFactory::instance().makeHeuristicNonOverlapping()
};
vector<double> storageOverheadThresholds =
{ 0.0, 0.25, 0.5, 0.75, 1.0, 1.25, 1.5, 1.75, 2.0 };
double total = solvers.size()
* storageOverheadThresholds.size()
* numRuns;
double completed = 0;
vector<util::RunningStat> io;
vector<util::RunningStat> storage;
vector<util::RunningStat> times;
vector<std::string> names;
vector<bool> errorFlags;
for (auto solver : solvers) {
io.push_back(util::RunningStat());
storage.push_back(util::RunningStat());
times.push_back(util::RunningStat());
names.push_back(solver->getClassName());
vector<std::string> names;
errorFlags.push_back(false);
}
int j;
for (double sot : storageOverheadThresholds) {
for (int i = 0; i < numRuns; i++) {
vector<std::unique_ptr<Attribute>> allAttributes;
auto workloadAndStats =
simConf.getQueryWorkloadAndStats(allAttributes);
QueryWorkload workload = workloadAndStats.first;
stats = workloadAndStats.second;
j = 0;
for (auto solver : solvers) {
timer.start();
Partitioning partitioning;
try {
partitioning = solver->solve(workload, sot, stats);
} catch (const runtime_error& error) {
cerr << "Unable to find a solution" << endl;
errorFlags.at(j) = true;
}
timer.stop();
if (!errorFlags.at(j)) {
io.at(j).push(cost.getIOCost(partitioning, workload));
storage.at(j).push(
cost.getStorageOverhead(partitioning, workload));
times.at(j).push(
timer.getRealTimeInSeconds());
}
j++;
cerr << ".";
completed++;
}
}
int j = 0;
for (auto solver : solvers) {
if (!errorFlags.at(j)) {
runningTimeExp.addRecord();
runningTimeExp.setFieldValue("solver", solver->getClassName());
runningTimeExp.setFieldValue(
"storageOverheadThreshold",
boost::str(boost::format("%.2f") % sot));
runningTimeExp.setFieldValue("time", times.at(j).getMean());
runningTimeExp.setFieldValue(
"deviation", times.at(j).getStandardDeviation());
times.at(j).clear();
queryIOExp.addRecord();
queryIOExp.setFieldValue("solver", solver->getClassName());
queryIOExp.setFieldValue(
"storageOverheadThreshold",
boost::str(boost::format("%.2f") % sot));
queryIOExp.setFieldValue("io", io.at(j).getMean());
queryIOExp.setFieldValue(
"deviation", io.at(j).getStandardDeviation());
io.at(j).clear();
storageExp.addRecord();
storageExp.setFieldValue("solver", solver->getClassName());
storageExp.setFieldValue(
"storageOverheadThreshold",
boost::str(boost::format("%.2f") % sot));
storageExp.setFieldValue("storage", storage.at(j).getMean());
storageExp.setFieldValue(
"deviation", storage.at(j).getStandardDeviation());
storage.at(j).clear();
} else {
runningTimeExp.addRecord();
runningTimeExp.setFieldValue("solver", solver->getClassName());
runningTimeExp.setFieldValue(
"storageOverheadThreshold",
boost::str(boost::format("%.2f") % sot));
runningTimeExp.setFieldValue("time", "n/a");
runningTimeExp.setFieldValue("deviation", "n/a");
times.at(j).clear();
queryIOExp.addRecord();
queryIOExp.setFieldValue("solver", solver->getClassName());
queryIOExp.setFieldValue(
"storageOverheadThreshold",
boost::str(boost::format("%.2f") % sot));
queryIOExp.setFieldValue("io", "n/a");
queryIOExp.setFieldValue("deviation", "n/a");
io.at(j).clear();
storageExp.addRecord();
storageExp.setFieldValue("solver", solver->getClassName());
storageExp.setFieldValue(
"storageOverheadThreshold",
boost::str(boost::format("%.2f") % sot));
storageExp.setFieldValue("storage", "n/a");
storageExp.setFieldValue("deviation", "n/a");
storage.at(j).clear();
errorFlags.at(j) = false;
}
j++;
}
cerr << " (" << (completed / total) * 100 << "%)" << endl;
}
for (auto exp : expData)
exp->close();
};
void VsBlockSize::process()
{
SimulationConf simConf;
double storageOverheadThreshold = 1.0;
assert(graphs_.size() >= 1);
simConf.setAttributeCount(
graphs_[0]->getConf().getEdgeSchema().getAttributes().size());
std::vector<std::vector<core::FocusedIntervalQuery>> queries;
std::vector<SchemaStats> stats;
std::vector<QueryWorkload> workloads;
std::cout << "Generating workload..." << std::endl;
for (int i=0; i < numRuns_; i++) {
std::cout << " " << i << "/" << numRuns_ << std::endl;
std::vector<std::vector<std::string> > templates =
simConf.getQueryTemplates(graphs_[0].get());
std::vector<core::FocusedIntervalQuery> qs =
ExpSetupHelper::genQueries(templates,
queryZipfParam_,
numQueries_,
tsStart_,
tsEnd_,
vertices_);
ExpSetupHelper::runWorkload(graphs_[0].get(),qs);
SchemaStats ss = graphs_[0]->getSchemaStats();
std::map<BucketId,common::QueryWorkload> ws =
graphs_[0]->getWorkloads();
// Make sure everything is in one bucket
assert(ws.size() == 1);
QueryWorkload w = ws.begin()->second;
// (queries, stats, workload)
queries.push_back(qs);
stats.push_back(ss);
workloads.push_back(w);
graphs_[0]->resetWorkloads();
}
std::cout << "done." << std::endl;
ExperimentalData edgeIOCountExp("EdgeIOCountVsBlockSize");
ExperimentalData edgeWriteIOCountExp("EdgeWriteIOCountVsBlockSize");
ExperimentalData edgeReadIOCountExp("EdgeReadIOCountVsBlockSize");
auto expData =
{ &edgeIOCountExp, &edgeWriteIOCountExp, &edgeReadIOCountExp };
makeEdgeIOCountExp(&edgeIOCountExp);
makeEdgeWriteIOCountExp(&edgeWriteIOCountExp);
makeEdgeReadIOCountExp(&edgeReadIOCountExp);
for (auto exp : expData)
exp->open();
vector<util::RunningStat> edgeIO;
vector<util::RunningStat> edgeWriteIO;
vector<util::RunningStat> edgeReadIO;
vector<std::string> names;
vector< shared_ptr<Solver> > solvers =
{
SolverFactory::instance().makeSinglePartition(),
SolverFactory::instance().makeOptimalNonOverlapping(),
SolverFactory::instance().makeHeuristicNonOverlapping()
};
for (auto solver : solvers) {
edgeIO.push_back(util::RunningStat());
edgeWriteIO.push_back(util::RunningStat());
edgeReadIO.push_back(util::RunningStat());
names.push_back(solver->getClassName());
}
int solverIndex;
size_t prevEdgeIOCount;
size_t prevEdgeReadIOCount;
size_t prevEdgeWriteIOCount;
int x = 0;
int total = graphs_.size() * numRuns_ * solvers.size();
std::cout << "Running experiments..." << std::endl;
int blockSizeIndex = -1;
for (auto iter = graphs_.begin(); iter != graphs_.end(); ++iter) {
blockSizeIndex++;
for (int i = 0; i < numRuns_; i++) {
solverIndex = -1;
for (auto solver : solvers) {
solverIndex++;
auto & partIndex = (*iter)->getPartitionIndex();
auto origParting =
partIndex.getTimeSlicedPartitioning(Timestamp(0.0));
intergdb::common::Partitioning solverSolution =
solver->solve(workloads[i], storageOverheadThreshold,
stats[i]);
std::cout << "Workload: "
<< workloads[i].toString() << std::endl;
std::cout << "Summary size: "
<< workloads[i].getQuerySummaries().size() << std::endl;
/*for (auto summary : workloads[i].getQuerySummaries())
std::cout << "Summary: "
<< summary.toString() << std::endl;
*/
std::cout << "Solver: " << solver->getClassName() << std::endl;
std::cout << solverSolution.toString() << std::endl;
TimeSlicedPartitioning newParting{}; // -inf to inf
newParting.getPartitioning() = solverSolution.toStringSet();
partIndex.replaceTimeSlicedPartitioning(
origParting, {newParting});
// to flush the filesystem cache
//system(“purge”);
prevEdgeIOCount = (*iter)->getEdgeIOCount();
prevEdgeReadIOCount = (*iter)->getEdgeReadIOCount();
prevEdgeWriteIOCount = (*iter)->getEdgeWriteIOCount();
ExpSetupHelper::runWorkload((*iter).get(),queries[i]);
std::cout <<
(*iter)->getEdgeIOCount() - prevEdgeIOCount << std::endl;
std::cout <<
(*iter)->getEdgeReadIOCount() - prevEdgeReadIOCount
<< std::endl;
std::cout <<
(*iter)->getEdgeWriteIOCount() - prevEdgeWriteIOCount
<< std::endl;
edgeIO[solverIndex].push(
(*iter)->getEdgeIOCount() - prevEdgeIOCount);
edgeReadIO[solverIndex].push(
(*iter)->getEdgeReadIOCount() - prevEdgeReadIOCount);
edgeWriteIO[solverIndex].push(
(*iter)->getEdgeWriteIOCount() - prevEdgeWriteIOCount);
x++;
std::cout << " " << x << "/" << total << std::endl;
}
}
for (int solverIndex = 0; solverIndex < solvers.size(); solverIndex++)
{
edgeIOCountExp.addRecord();
edgeIOCountExp.setFieldValue(
"solver", solvers[solverIndex]->getClassName());
edgeIOCountExp.setFieldValue(
"blockSize",
boost::lexical_cast<std::string>(blockSizes_[blockSizeIndex]));
edgeIOCountExp.setFieldValue(
"edgeIO", edgeIO[solverIndex].getMean());
edgeIOCountExp.setFieldValue(
"deviation", edgeIO[solverIndex].getStandardDeviation());
edgeIO[solverIndex].clear();
edgeWriteIOCountExp.addRecord();
edgeWriteIOCountExp.setFieldValue(
"solver", solvers[solverIndex]->getClassName());
edgeWriteIOCountExp.setFieldValue("blockSize",
boost::lexical_cast<std::string>(blockSizes_[blockSizeIndex]));
edgeWriteIOCountExp.setFieldValue(
"edgeWriteIO", edgeWriteIO[solverIndex].getMean());
edgeWriteIOCountExp.setFieldValue(
"deviation", edgeWriteIO[solverIndex].getStandardDeviation());
edgeWriteIO[solverIndex].clear();
edgeReadIOCountExp.addRecord();
edgeReadIOCountExp.setFieldValue(
"solver", solvers[solverIndex]->getClassName());
edgeReadIOCountExp.setFieldValue(
"blockSize",
boost::lexical_cast<std::string>(blockSizes_[blockSizeIndex]));
edgeReadIOCountExp.setFieldValue(
"edgeReadIO", edgeReadIO[solverIndex].getMean());
edgeReadIOCountExp.setFieldValue(
"deviation", edgeReadIO[solverIndex].getStandardDeviation());
edgeReadIO[solverIndex].clear();
}
}
std::cout << "done." << std::endl;
for (auto exp : expData)
exp->close();
};
void VsTimeDeltaBFS::process()
{
SimulationConf simConf;
double storageOverheadThreshold = 1.0;
assert(graph_ != NULL);
simConf.setAttributeCount(
graph_->getConf().getEdgeSchema().getAttributes().size());
simConf.setQueryTypeCount(numQueryTemplates_);
ExperimentalData edgeIOCountExp("EdgeIOCountVsTimeDeltaBFS");
ExperimentalData edgeWriteIOCountExp("EdgeWriteIOCountVsTimeDeltaBFS");
ExperimentalData edgeReadIOCountExp("EdgeReadIOCountVsTimeDeltaBFS");
auto expData =
{ &edgeIOCountExp, &edgeWriteIOCountExp, &edgeReadIOCountExp };
makeEdgeIOCountExp(&edgeIOCountExp);
makeEdgeWriteIOCountExp(&edgeWriteIOCountExp);
makeEdgeReadIOCountExp(&edgeReadIOCountExp);
for (auto exp : expData)
exp->open();
vector<util::RunningStat> edgeIO;
vector<util::RunningStat> edgeWriteIO;
vector<util::RunningStat> edgeReadIO;
vector<std::string> names;
vector< shared_ptr<Solver> > solvers =
{
SolverFactory::instance().makeSinglePartition(),
SolverFactory::instance().makeOptimalNonOverlapping(),
SolverFactory::instance().makeHeuristicNonOverlapping()
};
for (auto solver : solvers) {
edgeIO.push_back(util::RunningStat());
edgeWriteIO.push_back(util::RunningStat());
edgeReadIO.push_back(util::RunningStat());
names.push_back(solver->getClassName());
}
int solverIndex;
size_t prevEdgeIOCount;
size_t prevEdgeReadIOCount;
size_t prevEdgeWriteIOCount;
std::cout << "Running experiments..." << std::endl;
int deltaIndex = -1;
SchemaStats stats = graph_->getSchemaStats();
std::cout << stats.toString() << std::endl;
for (auto delta : timeDeltas_) {
deltaIndex++;
for (int i = 0; i < numRuns_; i++) {
// For each run with a different delta, generate a different set of queries:
// generate a different workload with numQueryTemplates
std::vector<std::vector<std::string> > templates =
simConf.getQueryTemplates(graph_.get());
std::vector<core::FocusedIntervalQuery> queries =
ExpSetupHelper::genSearchQueries(templates,
queryZipfParam_,
numQueries_,
tsStart_,
tsEnd_,
delta,
vertices_);
graph_->resetWorkloads();
ExpSetupHelper::runBFS(graph_.get(), queries);
std::map<BucketId,common::QueryWorkload> ws =
graph_->getWorkloads();
// Make sure everything is in one bucket
assert(ws.size() == 1);
QueryWorkload workload = ws.begin()->second;
solverIndex = -1;
for (auto solver : solvers) {
solverIndex++;
auto & partIndex = graph_->getPartitionIndex();
auto origParting =
partIndex.getTimeSlicedPartitioning(Timestamp(0.0));
intergdb::common::Partitioning solverSolution =
solver->solve(workload, storageOverheadThreshold, stats);
std::cout << "Solver: " << solver->getClassName() << std::endl;
std::cout << "numRuns: " << i << std::endl;
std::cout << "delta: " << delta << std::endl;
std::cout << solverSolution.toString() << std::endl;
TimeSlicedPartitioning newParting{}; // -inf to inf
newParting.getPartitioning() = solverSolution.toStringSet();
partIndex.replaceTimeSlicedPartitioning(
origParting, {newParting});
prevEdgeIOCount = graph_->getEdgeIOCount();
prevEdgeReadIOCount = graph_->getEdgeReadIOCount();
prevEdgeWriteIOCount = graph_->getEdgeWriteIOCount();
ExpSetupHelper::runBFS(graph_.get(),queries);
std::cout << "getEdgeIOCount: " <<
graph_->getEdgeIOCount() - prevEdgeIOCount << std::endl;
std::cout << "getEdgeReadIOCount: " <<
graph_->getEdgeReadIOCount() - prevEdgeReadIOCount
<< std::endl;
std::cout << "getEdgeWriteIOCount: " <<
graph_->getEdgeWriteIOCount() - prevEdgeWriteIOCount
<< std::endl;
edgeIO[solverIndex].push(
graph_->getEdgeIOCount() - prevEdgeIOCount);
edgeReadIO[solverIndex].push(
graph_->getEdgeReadIOCount() - prevEdgeReadIOCount);
edgeWriteIO[solverIndex].push(
graph_->getEdgeWriteIOCount() - prevEdgeWriteIOCount);
}
}
for (int solverIndex = 0; solverIndex < solvers.size(); solverIndex++)
{
edgeIOCountExp.addRecord();
edgeIOCountExp.setFieldValue(
"solver", solvers[solverIndex]->getClassName());
edgeIOCountExp.setFieldValue(
"delta",
boost::lexical_cast<std::string>(timeDeltas_[deltaIndex]));
edgeIOCountExp.setFieldValue(
"edgeIO", edgeIO[solverIndex].getMean());
edgeIOCountExp.setFieldValue(
"deviation", edgeIO[solverIndex].getStandardDeviation());
edgeIO[solverIndex].clear();
edgeWriteIOCountExp.addRecord();
edgeWriteIOCountExp.setFieldValue(
"solver", solvers[solverIndex]->getClassName());
edgeWriteIOCountExp.setFieldValue("delta",
boost::lexical_cast<std::string>(timeDeltas_[deltaIndex]));
edgeWriteIOCountExp.setFieldValue(
"edgeWriteIO", edgeWriteIO[solverIndex].getMean());
edgeWriteIOCountExp.setFieldValue(
"deviation", edgeWriteIO[solverIndex].getStandardDeviation());
edgeWriteIO[solverIndex].clear();
edgeReadIOCountExp.addRecord();
edgeReadIOCountExp.setFieldValue(
"solver", solvers[solverIndex]->getClassName());
edgeReadIOCountExp.setFieldValue(
"delta",
boost::lexical_cast<std::string>(timeDeltas_[deltaIndex]));
edgeReadIOCountExp.setFieldValue(
"edgeReadIO", edgeReadIO[solverIndex].getMean());
edgeReadIOCountExp.setFieldValue(
"deviation", edgeReadIO[solverIndex].getStandardDeviation());
edgeReadIO[solverIndex].clear();
}
}
// std::cout << "done." << std::endl;
for (auto exp : expData)
exp->close();
};
void VsNumAttributes::process()
{
cerr << "This is an experiment with name: "
<< this->getClassName() << endl;
SimulationConf simConf;
SchemaStats stats;
Cost cost(stats);
util::AutoTimer timer;
ExperimentalData queryIOExp("QueryIOVsNumAttributes");
ExperimentalData runningTimeExp("RunningTimeVsNumAttributes");
ExperimentalData storageExp("StorageOverheadVsNumAttributes");
auto expData = { &queryIOExp, &runningTimeExp, &storageExp };
makeQueryIOExp(&queryIOExp);
makeRunningTimeExp(&runningTimeExp);
makeStorageExp(&storageExp);
for (auto exp : expData)
exp->open();
auto solvers = {
SolverFactory::instance().makeSinglePartition(),
SolverFactory::instance().makePartitionPerAttribute(),
SolverFactory::instance().makeOptimalOverlapping(),
SolverFactory::instance().makeOptimalNonOverlapping(),
SolverFactory::instance().makeHeuristicOverlapping(),
SolverFactory::instance().makeHeuristicNonOverlapping()
};
//auto attributeCounts = {2, 4, 6, 8, 10, 12, 14, 16 };
auto attributeCounts = {32, 48, 64, 80, 96, 112, 128};
double total = solvers.size() * attributeCounts.size() * numRuns;
double completed = 0;
vector<util::RunningStat> io;
vector<util::RunningStat> storage;
vector<util::RunningStat> times;
vector<std::string> names;
for (auto solver : solvers) {
io.push_back(util::RunningStat());
storage.push_back(util::RunningStat());
times.push_back(util::RunningStat());
names.push_back(solver->getClassName());
vector<std::string> names;
}
int j;
for (double attributeCount : attributeCounts) {
for (int i = 0; i < numRuns; i++) {
simConf.setAttributeCount(attributeCount);
vector<unique_ptr<Attribute>> allAttributes;
auto workloadAndStats =
simConf.getQueryWorkloadAndStats(allAttributes);
QueryWorkload const& workload = workloadAndStats.first;
stats = workloadAndStats.second;
j = 0;
for (auto solver : solvers) {
timer.start();
Partitioning partitioning = solver->solve(
workload, storageOverheadThreshold, stats);
timer.stop();
io.at(j).push(cost.getIOCost(partitioning, workload));
storage.at(j).push(cost.getStorageOverhead(
partitioning, workload));
times.at(j).push(timer.getRealTimeInSeconds());
j++;
cerr << ".";
completed++;
}
}
int j = 0;
for (auto solver : solvers) {
runningTimeExp.addRecord();
runningTimeExp.setFieldValue("solver", solver->getClassName());
runningTimeExp.setFieldValue("attributes", attributeCount);
runningTimeExp.setFieldValue("time", times.at(j).getMean());
runningTimeExp.setFieldValue(
"deviation", times.at(j).getStandardDeviation());
times.at(j).clear();
queryIOExp.addRecord();
queryIOExp.setFieldValue("solver", solver->getClassName());
queryIOExp.setFieldValue("attributes", attributeCount);
queryIOExp.setFieldValue("io", io.at(j).getMean());
queryIOExp.setFieldValue(
"deviation", io.at(j).getStandardDeviation());
io.at(j).clear();
storageExp.addRecord();
storageExp.setFieldValue("solver", solver->getClassName());
storageExp.setFieldValue("attributes",attributeCount);
storageExp.setFieldValue("storage", storage.at(j).getMean());
storageExp.setFieldValue(
"deviation", storage.at(j).getStandardDeviation());
storage.at(j).clear();
j++;
}
cerr << " (" << (completed / total) * 100 << "%)" << endl;
}
for (auto exp : expData)
exp->close();
};
