TOOLS-2525: Improve performance of async benchmarks

.editorconfig

@@ -0,0 +1,12 @@
+root = true
+
+[*]
+charset = utf-8
+end_of_line = lf
+insert_final_newline = true
+trim_trailing_whitespace = true
+indent_style = tab
+indent_size = 4
+
+[*.{py,sh}]
+indent_style = space

src/include/benchmark.h

@@ -160,12 +160,18 @@ typedef struct threaddata_s {
 	struct thr_coordinator_s* coord;
 	as_random* random;
 	dyn_throttle_t dyn_throttle;
+	uint64_t min_usleep;
+	_Atomic(int64_t) async_req_quota;
+
 
 	// thread index: [0, n_threads)
 	uint32_t t_idx;
 	// which workload stage we're currrently on
 	_Atomic(uint32_t) stage_idx;
 
+	// For async linear workloads
+	_Atomic(uint64_t) current_key;
+
 	/*
 	 * note: to stop threads, tdata->finished must be set before tdata->do_work
 	 * to prevent deadlocking

src/main/benchmark.c

@@ -322,6 +322,7 @@ init_tdata(const args_t* args, cdata_t* cdata, thr_coord_t* coord,
 	tdata->t_idx = t_idx;
 	// always start on the first stage
 	atomic_init(&tdata->stage_idx, 0);
+	atomic_init(&tdata->current_key, args->start_key);
 
 	atomic_init(&tdata->do_work, true);
 	atomic_init(&tdata->finished, false);

src/main/benchmark_init.c

@@ -544,6 +544,8 @@ print_usage(const char* program)
 
 	printf("-z --threads <count> # Default: 16\n");
 	printf("   Load generating thread count.\n");
+	printf("   This is set to 1 if using --async.\n");
+	printf("   Use --event-loops in async mode.\n");
 	printf("\n");
 
 	printf("-g --throughput <tps> # Default: 0\n");
@@ -745,6 +747,7 @@ print_usage(const char* program)
 
 	printf("-a --async # Default: synchronous mode\n");
 	printf("   Enable asynchronous mode.\n");
+	printf("   Use --event-loops to tune performance in async mode.\n");
 	printf("\n");
 
 	printf("-c --async-max-commands <command count> # Default: 50\n");