diff --git a/stress-lockf.c b/stress-lockf.c
index fb2f7b5f085e..232c269d0ec8 100644
--- a/stress-lockf.c
+++ b/stress-lockf.c
@@ -43,7 +43,7 @@
 
 #define LOCK_FILE_SIZE	(64 * 1024)
 #define LOCK_SIZE	(8)
-#define LOCK_MAX	(1024)
+#define LOCK_MAX	(8192)
 
 typedef struct lockf_info {
 	off_t	offset;
diff --git a/stress-ng.c b/stress-ng.c
index 7ab2ab1c42a6..de48d2a34738 100644
--- a/stress-ng.c
+++ b/stress-ng.c
@@ -1642,12 +1642,22 @@ static void kill_procs(const int sig)
 		int j;
 
 		for (j = 0; j < procs[i].started_procs; j++) {
-			if (procs[i].pids[j])
-				(void)kill(procs[i].pids[j], signum);
+			if (procs[i].pids[j].pid)
+				(void)kill(procs[i].pids[j].pid, signum);
 		}
 	}
 }
 
+static int on_one_sock(unsigned cpu1, unsigned cpu2, unsigned cpus)
+{
+	unsigned mid = cpus / 2;
+	if (cpu1 < mid && cpu2 < mid)
+		return 1;
+	if (cpu1 >= mid && cpu2 >= mid)
+		return 1;
+	return 0;
+}
+
 /*
  *  wait_procs()
  * 	wait for procs
@@ -1670,7 +1680,8 @@ static void MLOCKED wait_procs(bool *success, bool *resource_success)
 		cpu_set_t proc_mask;
 		unsigned long int cpu = 0;
 		const uint32_t ticks_per_sec = stress_get_ticks_per_second() * 5;
-		const useconds_t usec_sleep = ticks_per_sec ? 1000000 / ticks_per_sec : 1000000 / 250;
+		//const useconds_t usec_sleep = ticks_per_sec ? 1000000 / ticks_per_sec : 1000000 / 250;
+		const useconds_t usec_sleep = 50000;
 
 		while (opt_do_wait) {
 			const int32_t cpus = stress_get_processors_configured();
@@ -1685,7 +1696,9 @@ static void MLOCKED wait_procs(bool *success, bool *resource_success)
 				int j;
 
 				for (j = 0; j < procs[i].started_procs; j++) {
-					const pid_t pid = procs[i].pids[j];
+					const pid_t pid = procs[i].pids[j].pid;
+					unsigned last_cpu =
+						procs[i].pids[j].last_cpu;
 					if (pid) {
 						cpu_set_t mask;
 						int32_t cpu_num;
@@ -1694,10 +1707,16 @@ static void MLOCKED wait_procs(bool *success, bool *resource_success)
 							cpu_num = mwc32() % cpus;
 						} while (!(CPU_ISSET(cpu_num, &proc_mask)));
 
+						if (on_one_sock(last_cpu,
+									cpu_num,
+									cpus))
+							cpu_num = cpus - cpu_num -1;
+
 						CPU_ZERO(&mask);
 						CPU_SET(cpu_num, &mask);
 						if (sched_setaffinity(pid, sizeof(mask), &mask) < 0)
 							goto do_wait;
+						procs[i].pids[j].last_cpu = cpu_num;
 					}
 				}
 			}
@@ -1713,7 +1732,7 @@ do_wait:
 		for (j = 0; j < procs[i].started_procs; j++) {
 			pid_t pid;
 redo:
-			pid = procs[i].pids[j];
+			pid = procs[i].pids[j].pid;
 			if (pid) {
 				int status, ret;
 
@@ -1752,7 +1771,7 @@ redo:
 						*success = false;
 						break;
 					}
-					proc_finished(&procs[i].pids[j]);
+					proc_finished(&procs[i].pids[j].pid);
 					pr_dbg(stderr, "process [%d] terminated\n", ret);
 				} else if (ret == -1) {
 					/* Somebody interrupted the wait */
@@ -1760,7 +1779,7 @@ redo:
 						goto redo;
 					/* This child did not exist, mark it done anyhow */
 					if (errno == ECHILD)
-						proc_finished(&procs[i].pids[j]);
+						proc_finished(&procs[i].pids[j].pid);
 				}
 			}
 		}
@@ -1901,7 +1920,7 @@ again:
 				default:
 					if (pid > -1) {
 						(void)setpgid(pid, pgrp);
-						procs[i].pids[j] = pid;
+						procs[i].pids[j].pid = pid;
 						procs[i].started_procs++;
 					}
 
@@ -2952,7 +2971,7 @@ next_opt:
 		/* Sequential mode has no bogo ops threshold */
 		for (i = 0; i < STRESS_MAX; i++) {
 			procs[i].bogo_ops = 0;
-			procs[i].pids = calloc(opt_sequential, sizeof(pid_t));
+			procs[i].pids = calloc(opt_sequential, sizeof(my_pid_t));
 			if (!procs[i].pids) {
 				pr_err(stderr, "cannot allocate pid list\n");
 				free_procs();
@@ -2980,7 +2999,7 @@ next_opt:
 			if (max_procs < procs[i].num_procs)
 				max_procs = procs[i].num_procs;
 			if (procs[i].num_procs) {
-				procs[i].pids = calloc(procs[i].num_procs, sizeof(pid_t));
+				procs[i].pids = calloc(procs[i].num_procs, sizeof(my_pid_t));
 				if (!procs[i].pids) {
 					pr_err(stderr, "cannot allocate pid list\n");
 					free_procs();
@@ -3008,7 +3027,7 @@ next_opt:
 			if (max_procs < procs[i].num_procs)
 				max_procs = procs[i].num_procs;
 			if (procs[i].num_procs) {
-				procs[i].pids = calloc(procs[i].num_procs, sizeof(pid_t));
+				procs[i].pids = calloc(procs[i].num_procs, sizeof(my_pid_t));
 				if (!procs[i].pids) {
 					pr_err(stderr, "cannot allocate pid list\n");
 					free_procs();
diff --git a/stress-ng.h b/stress-ng.h
index c5104a322e45..4f4098fc0d11 100644
--- a/stress-ng.h
+++ b/stress-ng.h
@@ -1713,7 +1713,12 @@ typedef struct {
 } stress_t;
 
 typedef struct {
-	pid_t	*pids;			/* process id */
+	pid_t pid;
+	unsigned last_cpu;
+} my_pid_t;
+
+typedef struct {
+	my_pid_t *pids;			/* process id */
 	int32_t started_procs;		/* count of started processes */
 	int32_t num_procs;		/* number of process per stressor */
 	uint64_t bogo_ops;		/* number of bogo ops */