diff --git a/test/e2e/framework/metrics_util.go b/test/e2e/framework/metrics_util.go
index 284beafde2..3b6b95c434 100644
--- a/test/e2e/framework/metrics_util.go
+++ b/test/e2e/framework/metrics_util.go
@@ -43,7 +43,6 @@ const (
 	// NodeStartupThreshold is a rough estimate of the time allocated for a pod to start on a node.
 	NodeStartupThreshold = 4 * time.Second
 
-	podStartupThreshold time.Duration = 5 * time.Second
 	// We are setting 1s threshold for apicalls even in small clusters to avoid flakes.
 	// The problem is that if long GC is happening in small clusters (where we have e.g.
 	// 1-core master machines) and tests are pretty short, it may consume significant
@@ -402,17 +401,17 @@ func HighLatencyRequests(c clientset.Interface, nodeCount int) (int, *APIRespons
 	return badMetrics, metrics, nil
 }
 
-// Verifies whether 50, 90 and 99th percentiles of e2e PodStartupLatency are
-// within the threshold.
-func VerifyPodStartupLatency(latency *PodStartupLatency) error {
-	if latency.E2ELatency.Perc50 > podStartupThreshold {
-		return fmt.Errorf("too high pod startup latency 50th percentile: %v", latency.E2ELatency.Perc50)
+// Verifies whether 50, 90 and 99th percentiles of a latency metric are
+// within the expected threshold.
+func VerifyLatencyWithinThreshold(threshold, actual LatencyMetric, metricName string) error {
+	if actual.Perc50 > threshold.Perc50 {
+		return fmt.Errorf("too high %v latency 50th percentile: %v", metricName, actual.Perc50)
 	}
-	if latency.E2ELatency.Perc90 > podStartupThreshold {
-		return fmt.Errorf("too high pod startup latency 90th percentile: %v", latency.E2ELatency.Perc90)
+	if actual.Perc90 > threshold.Perc90 {
+		return fmt.Errorf("too high %v latency 90th percentile: %v", metricName, actual.Perc90)
 	}
-	if latency.E2ELatency.Perc99 > podStartupThreshold {
-		return fmt.Errorf("too high pod startup latency 99th percentile: %v", latency.E2ELatency.Perc99)
+	if actual.Perc99 > threshold.Perc99 {
+		return fmt.Errorf("too high %v latency 99th percentile: %v", metricName, actual.Perc99)
 	}
 	return nil
 }
diff --git a/test/e2e/scalability/density.go b/test/e2e/scalability/density.go
index 553bc7a316..4f0c197934 100644
--- a/test/e2e/scalability/density.go
+++ b/test/e2e/scalability/density.go
@@ -50,6 +50,7 @@ import (
 )
 
 const (
+	PodStartupLatencyThreshold = 5 * time.Second
 	MinSaturationThreshold     = 2 * time.Minute
 	MinPodsPerSecondThroughput = 8
 	DensityPollInterval        = 10 * time.Second
@@ -846,7 +847,12 @@ var _ = SIGDescribe("Density", func() {
 				f.TestSummaries = append(f.TestSummaries, podStartupLatency)
 
 				// Test whether e2e pod startup time is acceptable.
-				framework.ExpectNoError(framework.VerifyPodStartupLatency(podStartupLatency))
+				podStartupLatencyThreshold := framework.LatencyMetric{
+					Perc50: PodStartupLatencyThreshold,
+					Perc90: PodStartupLatencyThreshold,
+					Perc99: PodStartupLatencyThreshold,
+				}
+				framework.ExpectNoError(framework.VerifyLatencyWithinThreshold(podStartupLatencyThreshold, podStartupLatency.E2ELatency, "pod startup"))
 
 				framework.LogSuspiciousLatency(startupLag, e2eLag, nodeCount, c)
 				latencyMeasurementPhase.End()
diff --git a/test/e2e_node/density_test.go b/test/e2e_node/density_test.go
index 845fff8cfe..9ee85cbe1d 100644
--- a/test/e2e_node/density_test.go
+++ b/test/e2e_node/density_test.go
@@ -472,21 +472,6 @@ func getPodStartLatency(node string) (framework.KubeletLatencyMetrics, error) {
 	return latencyMetrics, nil
 }
 
-// verifyPodStartupLatency verifies whether 50, 90 and 99th percentiles of PodStartupLatency are
-// within the threshold.
-func verifyPodStartupLatency(expect, actual framework.LatencyMetric) error {
-	if actual.Perc50 > expect.Perc50 {
-		return fmt.Errorf("too high pod startup latency 50th percentile: %v", actual.Perc50)
-	}
-	if actual.Perc90 > expect.Perc90 {
-		return fmt.Errorf("too high pod startup latency 90th percentile: %v", actual.Perc90)
-	}
-	if actual.Perc99 > expect.Perc99 {
-		return fmt.Errorf("too high pod startup latency 99th percentile: %v", actual.Perc99)
-	}
-	return nil
-}
-
 // newInformerWatchPod creates an informer to check whether all pods are running.
 func newInformerWatchPod(f *framework.Framework, mutex *sync.Mutex, watchTimes map[string]metav1.Time, podType string) cache.Controller {
 	ns := f.Namespace.Name
@@ -563,7 +548,7 @@ func logAndVerifyLatency(batchLag time.Duration, e2eLags []framework.PodLatencyD
 
 	if isVerify {
 		// check whether e2e pod startup time is acceptable.
-		framework.ExpectNoError(verifyPodStartupLatency(podStartupLimits, podStartupLatency))
+		framework.ExpectNoError(framework.VerifyLatencyWithinThreshold(podStartupLimits, podStartupLatency, "pod startup"))
 
 		// check bactch pod creation latency
 		if podBatchStartupLimit > 0 {