diff --git a/slo-monitor/Makefile b/slo-monitor/Makefile
index ed22e34481..52ff25c162 100644
--- a/slo-monitor/Makefile
+++ b/slo-monitor/Makefile
@@ -13,7 +13,7 @@
 # limitations under the License.
 
 PACKAGE = k8s.io/perf-tests/slo-monitor
-TAG = 0.14.0
+TAG = 0.15.0
 # Image should be pulled from k8s.gcr.io, which will auto-detect
 # region (us, eu, asia, ...) and pull from the closest.
 REPOSITORY?=k8s.gcr.io
diff --git a/slo-monitor/src/monitors/pod_monitor.go b/slo-monitor/src/monitors/pod_monitor.go
index 6be34c1949..10fc64139d 100644
--- a/slo-monitor/src/monitors/pod_monitor.go
+++ b/slo-monitor/src/monitors/pod_monitor.go
@@ -129,8 +129,20 @@ func (pm *PodStartupLatencyDataMonitor) Run(stopCh chan struct{}) error {
 	controller := NewWatcherWithHandler(
 		&cache.ListWatch{
 			ListFunc: func(options metav1.ListOptions) (runtime.Object, error) {
-				// This call is paginated by cache.Reflector.
-				return pm.kubeClient.CoreV1().Pods(v1.NamespaceAll).List(options)
+				// slo-monitor is not using the events returned from the list
+				// (it uses NoStoreQueue implementation of Store which discards them),
+				// only the resourceVersion is used to instantiate the watch from this point.
+				// This trick allows us to reduce memory usage on startup and further relists.
+				o := metav1.ListOptions{
+					Limit: 1,
+				}
+				result, err := pm.kubeClient.CoreV1().Pods(v1.NamespaceAll).List(o)
+				if err != nil {
+					return nil, err
+				}
+				result.Continue = ""
+				result.Items = nil
+				return result, nil
 			},
 			WatchFunc: func(options metav1.ListOptions) (watch.Interface, error) {
 				return pm.kubeClient.CoreV1().Pods(v1.NamespaceAll).Watch(options)
@@ -175,9 +187,20 @@ func (pm *PodStartupLatencyDataMonitor) Run(stopCh chan struct{}) error {
 	eventController := NewWatcherWithHandler(
 		&cache.ListWatch{
 			ListFunc: func(options metav1.ListOptions) (runtime.Object, error) {
-				options.FieldSelector = eventSelector
-				// This call is paginated by cache.Reflector.
-				return pm.kubeClient.CoreV1().Events(v1.NamespaceAll).List(options)
+				// slo-monitor is not using the pods returned from the list
+				// (it uses NoStoreQueue implementation of Store which discards them),
+				// only the resourceVersion is used to instantiate the watch from this point.
+				// This trick allows us to reduce memory usage on startup and further relists.
+				o := metav1.ListOptions{
+					Limit: 1,
+				}
+				result, err := pm.kubeClient.CoreV1().Events(v1.NamespaceAll).List(o)
+				if err != nil {
+					return nil, err
+				}
+				result.Continue = ""
+				result.Items = nil
+				return result, nil
 			},
 			WatchFunc: func(options metav1.ListOptions) (watch.Interface, error) {
 				options.FieldSelector = eventSelector
diff --git a/slo-monitor/src/monitors/watcher.go b/slo-monitor/src/monitors/watcher.go
index c2a941a107..be36dfa6e6 100644
--- a/slo-monitor/src/monitors/watcher.go
+++ b/slo-monitor/src/monitors/watcher.go
@@ -28,7 +28,6 @@ import (
 
 const (
 	resyncPeriod      = time.Duration(0)
-	watchListPageSize = 5000
 )
 
 // NewWatcherWithHandler creates a simple watcher that will call `h` for all coming objects
@@ -41,7 +40,6 @@ func NewWatcherWithHandler(lw cache.ListerWatcher, objType runtime.Object, setHa
 		ObjectType:        objType,
 		FullResyncPeriod:  resyncPeriod,
 		RetryOnError:      false,
-		WatchListPageSize: watchListPageSize,
 
 		Process: func(obj interface{}) error {
 			workItem := obj.(workItem)