Implement calibratable XRootD computational overhead

data/workload-configs/crown_ttbar_copyjob.json

@@ -7,7 +7,7 @@
         },
         "flops": {
             "type": "gaussian",
-            "average": 1174794431328,
+            "average": 79613793103,
             "sigma": 0
         },
         "memory": {

src/SimpleSimulator.cpp

@@ -52,6 +52,7 @@ bool SimpleSimulator::infile_caching_on = true;                  // flag to turn
 bool SimpleSimulator::prefetching_on = true;                     // flag to enable prefetching during streaming
 bool SimpleSimulator::shuffle_jobs = false;                      // flag to enable job shuffling during submission
 double SimpleSimulator::xrd_block_size = 1. * 1000 * 1000 * 1000;// maximum size of the streamed file blocks in bytes for the XRootD-ish streaming
+double SimpleSimulator::xrd_add_flops_per_time = 20000000000;// flops overhead introduced by XRootD streaming per second
 // TODO: The initialized below is likely bogus (at compile time?)
 std::set<std::string> SimpleSimulator::cache_hosts;
 std::set<std::string> SimpleSimulator::storage_hosts;
@@ -243,6 +244,7 @@ po::variables_map process_program_options(int argc, char **argv) {
     bool shuffle_jobs = false;
 
     double xrd_block_size = 1000. * 1000 * 1000;
+    double xrd_add_flops_per_time = 20000000000;
     std::string storage_service_buffer_size = "1048576";// 1MiB
 
     unsigned int seed = 42;
@@ -256,6 +258,7 @@ po::variables_map process_program_options(int argc, char **argv) {
     op("no-caching", po::bool_switch()->default_value(no_caching), "switch to turn on/off the caching of jobs' input-files")("prefetch-off", po::bool_switch()->default_value(prefetch_off), "switch to turn on/off prefetching for streaming of input-files")("shuffle-jobs", po::bool_switch()->default_value(shuffle_jobs), "switch to turn on/off shuffling jobs during submission");
     op("output-file,o", po::value<std::string>()->value_name("<out file>")->required(), "path for the CSV file containing output information about the jobs in the simulation");
     op("xrd-blocksize,x", po::value<double>()->default_value(xrd_block_size), "size of the blocks XRootD uses for data streaming")("storage-buffer-size,b", po::value<StorageServiceBufferValue>()->default_value(StorageServiceBufferValue(storage_service_buffer_size)), "buffer size used by the storage services when communicating data");
+    op("xrd-flops-per-time", po::value<double>()->default_value(xrd_add_flops_per_time), "flops overhead introduced by XRootD data streaming per second");
     op("cache-scope", po::value<cacheScope>()->default_value(cacheScope("local")), "Set the network scope in which caches can be found:\n local: only caches on same machine\n network: caches in same network zone\n siblingnetwork: also include caches in sibling networks");
     op("seed,s", po::value<unsigned int>()->default_value(seed), "Set the seed for the random generator");
 
@@ -450,6 +453,7 @@ int main(int argc, char **argv) {
 
     // Set XRootD block size
     SimpleSimulator::xrd_block_size = vm["xrd-blocksize"].as<double>();
+    SimpleSimulator::xrd_add_flops_per_time = vm["xrd-flops-per-time"].as<double>();
 
     // Set StorageService buffer size/type
     std::string buffer_size = vm["storage-buffer-size"].as<StorageServiceBufferValue>().get();

src/SimpleSimulator.h

@@ -21,18 +21,22 @@ class SimpleSimulator {
     static std::set<std::string> network_monitors;// hosts configured to monitor network
 
     static void fillHostsInSiblingZonesMap(bool include_subzones);
-    static bool local_cache_scope;
 
     static std::map<std::string, std::set<std::string>> hosts_in_zones;// map holding information of all hosts present in network zones
+    static std::map<std::shared_ptr<wrench::StorageService>, LRU_FileList> global_file_map;// map holding files informations
 
+    // global simulator settings and parameters
     static bool infile_caching_on;
     static bool prefetching_on;
+    static bool local_cache_scope;
+
     static bool shuffle_jobs;
-    static std::map<std::shared_ptr<wrench::StorageService>, LRU_FileList> global_file_map;
+
     static double xrd_block_size;
+    static double xrd_add_flops_per_time;
     static std::mt19937 gen;
 
-    // Cores required
+    /*// Cores required
     static int req_cores;
     // Flops distribution
     static double mean_flops;
@@ -49,7 +53,7 @@ class SimpleSimulator {
     // Output-file distribution
     static double mean_outsize;
     static double sigma_outsize;
-    static std::normal_distribution<double> *outsize_dist;
+    static std::normal_distribution<double> *outsize_dist;*/
 
     /** @brief Output filestream object to write out dump */
     static std::ofstream filedump;

src/computation/StreamedComputation.cpp

@@ -36,6 +36,8 @@ void StreamedComputation::performComputation(std::shared_ptr<wrench::ActionExecu
 
     auto the_action = std::dynamic_pointer_cast<MonitorAction>(action_executor->getAction());// executed action
 
+    double job_start_time = wrench::Simulation::getCurrentSimulatedDate();
+
     double infile_transfer_time = 0.;
     double compute_time = 0.;
 
@@ -67,6 +69,10 @@ void StreamedComputation::performComputation(std::shared_ptr<wrench::ActionExecu
             double num_bytes = std::min<double>(SimpleSimulator::xrd_block_size, data_to_process);
             double num_flops = determineFlops(num_bytes, total_data_size);
             WRENCH_INFO("Chunk: %.2lf bytes / %.2lf flops", num_bytes, num_flops);
+            // Add XRootD FLOPs overhead that increments with execution time
+            double xrd_overhead_flops = SimpleSimulator::xrd_add_flops_per_time * (wrench::Simulation::getCurrentSimulatedDate() - job_start_time);
+            num_flops += xrd_overhead_flops;
+            WRENCH_DEBUG("       + %.2lf flops XRootD overhead", xrd_overhead_flops);
             // Start the computation asynchronously
             simgrid::s4u::ExecPtr exec = simgrid::s4u::this_actor::exec_init(num_flops);
             double exec_start_time = 0.0;

tools/hitratePerformancePlots.py

@@ -224,8 +224,9 @@ def mapHostToSite(test: str, mapping: 'dict[str,str]',):
             hue="Site", hue_order=sites,
             data=df,
             estimator="median", errorbar=("pi",95), # ci = Confidence Interval, pi = Percentile Interval, sd = Standard Deviation, se = Standard Error of Mean
-            dodge=True, join=False,
-            markers=markers, capsize=0.5/len(sites), errwidth=1.,
+            dodge=True,
+            linestyle='none',
+            markers=markers, capsize=0.5/len(sites), err_kws={'linewidth': 1.0},
             palette=sns.color_palette("colorblind", n_colors=len(sites)),
             ax=ax1
         )

tools/hitrateScan.sh

@@ -16,8 +16,8 @@ action() {
     local parent="$( dirname "$base" )"
 
     local PLATFORM="$parent/data/platform-files/sgbatch_validation.xml"
-    local WORKLOAD="$parent/data/workload-configs/crown_ttbar_validation.json"
-    local DATASET="$parent/data/dataset-configs/sample.json"
+    local WORKLOAD="$parent/data/workload-configs/crown_ttbar_copyjob.json"
+    local DATASET="$parent/data/dataset-configs/crown_ttbar_copyjob.json"
 
     # local NJOBS=1
     # local NINFILES=20 #10
@@ -32,28 +32,29 @@ action() {
 
     local DUPLICATIONS=48
 
-    local XRD_BLOCKSIZE=100000000
-    local STORAGE_BUFFER_SIZE=1048576
+    local XRD_BLOCKSIZE=10000000000
+    local STORAGE_BUFFER_SIZE=0
 
-    local SCENARIO="fullstream" # further options synchronized with plotting script "copy", "simplifiedstream", "fullstream"
+    local SCENARIO="fastNetworkfastCache" # further options synchronized with plotting script "copy", "simplifiedstream", "fullstream"
 
-    local OUTDIR="$parent/tmp/outputs"
+    local OUTDIR="$parent/tmp/outputs/copyjobs"
     if [ ! -d $OUTDIR ]; then
         mkdir -p $OUTDIR
     fi
 
     for hitrate in $(LANG=en_UK seq 0.0 0.1 1.0)
     do 
         dc-sim --platform "$PLATFORM" \
-            --hitrate ${hitrate} \
+            --workload-configurations "$WORKLOAD" \
+            --dataset-configurations "$DATASET" \
             --duplications $DUPLICATIONS \
+            --hitrate ${hitrate} \
             --xrd-blocksize $XRD_BLOCKSIZE \
-            --output-file ${OUTDIR}/hitratescaling_${SCENARIO}_xrd${XRD_BLOCKSIZE}_${NJOBS}jobs_hitrate${hitrate}.csv \
-            --cfg=network/loopback-bw:100000000000000 \
             --storage-buffer-size $STORAGE_BUFFER_SIZE \
+            --cfg=network/loopback-bw:100000000000000 \
             --no-caching \
-            --workload-configurations "$WORKLOAD" \
-            --dataset-configurations "$DATASET" #\
+            --seed 42 \
+            --output-file ${OUTDIR}/hitratescaling_${SCENARIO}_xrd${XRD_BLOCKSIZE}_${NJOBS}jobs_hitrate${hitrate}.csv
             # --no-streaming \
             # --wrench-full-log
             # --log=simple_wms.threshold=debug \