diff --git a/cmd/kubemark/hollow-node.go b/cmd/kubemark/hollow-node.go
index 86c0dd85511..b05f035b168 100644
--- a/cmd/kubemark/hollow-node.go
+++ b/cmd/kubemark/hollow-node.go
@@ -20,10 +20,11 @@ import (
 	"fmt"
 
 	"k8s.io/kubernetes/pkg/api"
+	"k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
 	_ "k8s.io/kubernetes/pkg/client/metrics/prometheus" // for client metric registration
 	"k8s.io/kubernetes/pkg/client/record"
+	"k8s.io/kubernetes/pkg/client/restclient"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
-	clientset "k8s.io/kubernetes/pkg/client/unversioned/adapters/internalclientset"
 	"k8s.io/kubernetes/pkg/client/unversioned/clientcmd"
 	cadvisortest "k8s.io/kubernetes/pkg/kubelet/cadvisor/testing"
 	"k8s.io/kubernetes/pkg/kubelet/cm"
@@ -66,7 +67,7 @@ func (c *HollowNodeConfig) addFlags(fs *pflag.FlagSet) {
 	fs.StringVar(&c.ContentType, "kube-api-content-type", "application/vnd.kubernetes.protobuf", "ContentType of requests sent to apiserver.")
 }
 
-func (c *HollowNodeConfig) createClientFromFile() (*client.Client, error) {
+func (c *HollowNodeConfig) createClientConfigFromFile() (*restclient.Config, error) {
 	clientConfig, err := clientcmd.LoadFromFile(c.KubeconfigPath)
 	if err != nil {
 		return nil, fmt.Errorf("error while loading kubeconfig from file %v: %v", c.KubeconfigPath, err)
@@ -76,15 +77,10 @@ func (c *HollowNodeConfig) createClientFromFile() (*client.Client, error) {
 		return nil, fmt.Errorf("error while creating kubeconfig: %v", err)
 	}
 	config.ContentType = c.ContentType
-	client, err := client.New(config)
-	if err != nil {
-		return nil, fmt.Errorf("error while creating client: %v", err)
-	}
-	return client, nil
+	return config, nil
 }
 
 func main() {
-
 	config := HollowNodeConfig{}
 	config.addFlags(pflag.CommandLine)
 	flag.InitFlags()
@@ -94,10 +90,17 @@ func main() {
 	}
 
 	// create a client to communicate with API server.
-	cl, err := config.createClientFromFile()
-	clientset := clientset.FromUnversionedClient(cl)
+	clientConfig, err := config.createClientConfigFromFile()
 	if err != nil {
-		glog.Fatal("Failed to create a Client. Exiting.")
+		glog.Fatalf("Failed to create a ClientConfig: %v. Exiting.", err)
+	}
+	cl, err := client.New(clientConfig)
+	if err != nil {
+		glog.Fatalf("Failed to create a Client: %v. Exiting.", err)
+	}
+	clientset, err := internalclientset.NewForConfig(clientConfig)
+	if err != nil {
+		glog.Fatalf("Failed to create a ClientSet: %v. Exiting.", err)
 	}
 
 	if config.Morph == "kubelet" {
diff --git a/pkg/client/unversioned/adapters/internalclientset/clientset_adaption.go b/pkg/client/unversioned/adapters/internalclientset/clientset_adaption.go
deleted file mode 100644
index b9b348d7777..00000000000
--- a/pkg/client/unversioned/adapters/internalclientset/clientset_adaption.go
+++ /dev/null
@@ -1,79 +0,0 @@
-/*
-Copyright 2016 The Kubernetes Authors.
-
-Licensed under the Apache License, Version 2.0 (the "License");
-you may not use this file except in compliance with the License.
-You may obtain a copy of the License at
-
-    http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software
-distributed under the License is distributed on an "AS IS" BASIS,
-WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-See the License for the specific language governing permissions and
-limitations under the License.
-*/
-
-package internalclientset
-
-import (
-	"k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
-	unversionedauthentication "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset/typed/authentication/unversioned"
-	unversionedauthorization "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset/typed/authorization/unversioned"
-	unversionedautoscaling "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset/typed/autoscaling/unversioned"
-	unversionedbatch "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset/typed/batch/unversioned"
-	unversionedcore "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset/typed/core/unversioned"
-	unversionedextensions "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset/typed/extensions/unversioned"
-	unversionedstorage "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset/typed/storage/unversioned"
-	"k8s.io/kubernetes/pkg/client/typed/discovery"
-	"k8s.io/kubernetes/pkg/client/unversioned"
-)
-
-// FromUnversionedClient adapts a unversioned.Client to a internalclientset.Clientset.
-// This function is temporary. We will remove it when everyone has moved to using
-// Clientset. New code should NOT use this function.
-func FromUnversionedClient(c *unversioned.Client) *internalclientset.Clientset {
-	var clientset internalclientset.Clientset
-	if c != nil {
-		clientset.CoreClient = unversionedcore.New(c.RESTClient)
-	} else {
-		clientset.CoreClient = unversionedcore.New(nil)
-	}
-	if c != nil && c.ExtensionsClient != nil {
-		clientset.ExtensionsClient = unversionedextensions.New(c.ExtensionsClient.RESTClient)
-	} else {
-		clientset.ExtensionsClient = unversionedextensions.New(nil)
-	}
-	if c != nil && c.BatchClient != nil {
-		clientset.BatchClient = unversionedbatch.New(c.BatchClient.RESTClient)
-	} else {
-		clientset.BatchClient = unversionedbatch.New(nil)
-	}
-	if c != nil && c.AuthorizationClient != nil {
-		clientset.AuthorizationClient = unversionedauthorization.New(c.AuthorizationClient.RESTClient)
-	} else {
-		clientset.AuthorizationClient = unversionedauthorization.New(nil)
-	}
-	if c != nil && c.AutoscalingClient != nil {
-		clientset.AutoscalingClient = unversionedautoscaling.New(c.AutoscalingClient.RESTClient)
-	} else {
-		clientset.AutoscalingClient = unversionedautoscaling.New(nil)
-	}
-	if c != nil && c.AuthenticationClient != nil {
-		clientset.AuthenticationClient = unversionedauthentication.New(c.AuthenticationClient.RESTClient)
-	} else {
-		clientset.AuthenticationClient = unversionedauthentication.New(nil)
-	}
-	if c != nil && c.DiscoveryClient != nil {
-		clientset.DiscoveryClient = discovery.NewDiscoveryClient(c.DiscoveryClient.RESTClient)
-	} else {
-		clientset.DiscoveryClient = discovery.NewDiscoveryClient(nil)
-	}
-	if c != nil && c.StorageClient != nil {
-		clientset.StorageClient = unversionedstorage.New(c.StorageClient.RESTClient)
-	} else {
-		clientset.StorageClient = unversionedstorage.New(nil)
-	}
-
-	return &clientset
-}
diff --git a/test/e2e/autoscaling_utils.go b/test/e2e/autoscaling_utils.go
index 603683d3594..2d64991b9e1 100644
--- a/test/e2e/autoscaling_utils.go
+++ b/test/e2e/autoscaling_utils.go
@@ -302,9 +302,9 @@ func (rc *ResourceConsumer) CleanUp() {
 	rc.stopCustomMetric <- 0
 	// Wait some time to ensure all child goroutines are finished.
 	time.Sleep(10 * time.Second)
-	framework.ExpectNoError(framework.DeleteRCAndPods(rc.framework.Client, rc.framework.Namespace.Name, rc.name))
+	framework.ExpectNoError(framework.DeleteRCAndPods(rc.framework.Client, rc.framework.ClientSet, rc.framework.Namespace.Name, rc.name))
 	framework.ExpectNoError(rc.framework.Client.Services(rc.framework.Namespace.Name).Delete(rc.name))
-	framework.ExpectNoError(framework.DeleteRCAndPods(rc.framework.Client, rc.framework.Namespace.Name, rc.controllerName))
+	framework.ExpectNoError(framework.DeleteRCAndPods(rc.framework.Client, rc.framework.ClientSet, rc.framework.Namespace.Name, rc.controllerName))
 	framework.ExpectNoError(rc.framework.Client.Services(rc.framework.Namespace.Name).Delete(rc.controllerName))
 }
 
diff --git a/test/e2e/batch_v1_jobs.go b/test/e2e/batch_v1_jobs.go
index a33d4924fd7..daa99e2aff4 100644
--- a/test/e2e/batch_v1_jobs.go
+++ b/test/e2e/batch_v1_jobs.go
@@ -27,7 +27,6 @@ import (
 	"k8s.io/kubernetes/pkg/api/errors"
 	"k8s.io/kubernetes/pkg/apis/batch"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
-	clientsetadapter "k8s.io/kubernetes/pkg/client/unversioned/adapters/internalclientset"
 	"k8s.io/kubernetes/pkg/kubectl"
 	"k8s.io/kubernetes/pkg/labels"
 	"k8s.io/kubernetes/pkg/util/wait"
@@ -129,7 +128,7 @@ var _ = framework.KubeDescribe("V1Job", func() {
 		Expect(err).NotTo(HaveOccurred())
 
 		By("scale job up")
-		scaler, err := kubectl.ScalerFor(batch.Kind("Job"), clientsetadapter.FromUnversionedClient(f.Client))
+		scaler, err := kubectl.ScalerFor(batch.Kind("Job"), f.ClientSet)
 		Expect(err).NotTo(HaveOccurred())
 		waitForScale := kubectl.NewRetryParams(5*time.Second, 1*time.Minute)
 		waitForReplicas := kubectl.NewRetryParams(5*time.Second, 5*time.Minute)
@@ -154,7 +153,7 @@ var _ = framework.KubeDescribe("V1Job", func() {
 		Expect(err).NotTo(HaveOccurred())
 
 		By("scale job down")
-		scaler, err := kubectl.ScalerFor(batch.Kind("Job"), clientsetadapter.FromUnversionedClient(f.Client))
+		scaler, err := kubectl.ScalerFor(batch.Kind("Job"), f.ClientSet)
 		Expect(err).NotTo(HaveOccurred())
 		waitForScale := kubectl.NewRetryParams(5*time.Second, 1*time.Minute)
 		waitForReplicas := kubectl.NewRetryParams(5*time.Second, 5*time.Minute)
@@ -177,7 +176,7 @@ var _ = framework.KubeDescribe("V1Job", func() {
 		Expect(err).NotTo(HaveOccurred())
 
 		By("delete a job")
-		reaper, err := kubectl.ReaperFor(batch.Kind("Job"), clientsetadapter.FromUnversionedClient(f.Client))
+		reaper, err := kubectl.ReaperFor(batch.Kind("Job"), f.ClientSet)
 		Expect(err).NotTo(HaveOccurred())
 		timeout := 1 * time.Minute
 		err = reaper.Stop(f.Namespace.Name, job.Name, timeout, api.NewDeleteOptions(0))
diff --git a/test/e2e/cluster_size_autoscaling.go b/test/e2e/cluster_size_autoscaling.go
index cbe9df69448..dc9d6c659e0 100644
--- a/test/e2e/cluster_size_autoscaling.go
+++ b/test/e2e/cluster_size_autoscaling.go
@@ -97,7 +97,7 @@ var _ = framework.KubeDescribe("Cluster size autoscaling [Slow]", func() {
 	It("shouldn't increase cluster size if pending pod is too large [Feature:ClusterSizeAutoscalingScaleUp]", func() {
 		By("Creating unschedulable pod")
 		ReserveMemory(f, "memory-reservation", 1, memCapacityMb, false)
-		defer framework.DeleteRCAndPods(f.Client, f.Namespace.Name, "memory-reservation")
+		defer framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, "memory-reservation")
 
 		By("Waiting for scale up hoping it won't happen")
 		// Verfiy, that the appropreate event was generated.
@@ -124,7 +124,7 @@ var _ = framework.KubeDescribe("Cluster size autoscaling [Slow]", func() {
 
 	It("should increase cluster size if pending pods are small [Feature:ClusterSizeAutoscalingScaleUp]", func() {
 		ReserveMemory(f, "memory-reservation", 100, nodeCount*memCapacityMb, false)
-		defer framework.DeleteRCAndPods(f.Client, f.Namespace.Name, "memory-reservation")
+		defer framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, "memory-reservation")
 
 		// Verify, that cluster size is increased
 		framework.ExpectNoError(WaitForClusterSizeFunc(f.Client,
@@ -143,7 +143,7 @@ var _ = framework.KubeDescribe("Cluster size autoscaling [Slow]", func() {
 		glog.Infof("Not enabling cluster autoscaler for the node pool (on purpose).")
 
 		ReserveMemory(f, "memory-reservation", 100, nodeCount*memCapacityMb, false)
-		defer framework.DeleteRCAndPods(f.Client, f.Namespace.Name, "memory-reservation")
+		defer framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, "memory-reservation")
 
 		// Verify, that cluster size is increased
 		framework.ExpectNoError(WaitForClusterSizeFunc(f.Client,
@@ -165,7 +165,7 @@ var _ = framework.KubeDescribe("Cluster size autoscaling [Slow]", func() {
 
 	It("should increase cluster size if pods are pending due to host port conflict [Feature:ClusterSizeAutoscalingScaleUp]", func() {
 		CreateHostPortPods(f, "host-port", nodeCount+2, false)
-		defer framework.DeleteRCAndPods(f.Client, f.Namespace.Name, "host-port")
+		defer framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, "host-port")
 
 		framework.ExpectNoError(WaitForClusterSizeFunc(f.Client,
 			func(size int) bool { return size >= nodeCount+2 }, scaleUpTimeout))
@@ -217,7 +217,7 @@ var _ = framework.KubeDescribe("Cluster size autoscaling [Slow]", func() {
 			func(size int) bool { return size >= nodeCount+1 }, scaleUpTimeout))
 
 		framework.ExpectNoError(waitForAllCaPodsReadyInNamespace(f, c))
-		framework.ExpectNoError(framework.DeleteRCAndPods(f.Client, f.Namespace.Name, "node-selector"))
+		framework.ExpectNoError(framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, "node-selector"))
 	})
 
 	It("should scale up correct target pool [Feature:ClusterSizeAutoscalingScaleUp]", func() {
@@ -232,7 +232,7 @@ var _ = framework.KubeDescribe("Cluster size autoscaling [Slow]", func() {
 
 		By("Creating rc with 2 pods too big to fit default-pool but fitting extra-pool")
 		ReserveMemory(f, "memory-reservation", 2, 2*memCapacityMb, false)
-		defer framework.DeleteRCAndPods(f.Client, f.Namespace.Name, "memory-reservation")
+		defer framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, "memory-reservation")
 
 		// Apparently GKE master is restarted couple minutes after the node pool is added
 		// reseting all the timers in scale down code. Adding 5 extra minutes to workaround
diff --git a/test/e2e/daemon_restart.go b/test/e2e/daemon_restart.go
index ad3a9d8a5f5..61d84b758c8 100644
--- a/test/e2e/daemon_restart.go
+++ b/test/e2e/daemon_restart.go
@@ -261,7 +261,7 @@ var _ = framework.KubeDescribe("DaemonRestart [Disruptive]", func() {
 		// that it had the opportunity to create/delete pods, if it were going to do so. Scaling the RC
 		// to the same size achieves this, because the scale operation advances the RC's sequence number
 		// and awaits it to be observed and reported back in the RC's status.
-		framework.ScaleRC(f.Client, ns, rcName, numPods, true)
+		framework.ScaleRC(f.Client, f.ClientSet, ns, rcName, numPods, true)
 
 		// Only check the keys, the pods can be different if the kubelet updated it.
 		// TODO: Can it really?
@@ -292,9 +292,9 @@ var _ = framework.KubeDescribe("DaemonRestart [Disruptive]", func() {
 		restarter.kill()
 		// This is best effort to try and create pods while the scheduler is down,
 		// since we don't know exactly when it is restarted after the kill signal.
-		framework.ExpectNoError(framework.ScaleRC(f.Client, ns, rcName, numPods+5, false))
+		framework.ExpectNoError(framework.ScaleRC(f.Client, f.ClientSet, ns, rcName, numPods+5, false))
 		restarter.waitUp()
-		framework.ExpectNoError(framework.ScaleRC(f.Client, ns, rcName, numPods+5, true))
+		framework.ExpectNoError(framework.ScaleRC(f.Client, f.ClientSet, ns, rcName, numPods+5, true))
 	})
 
 	It("Kubelet should not restart containers across restart", func() {
diff --git a/test/e2e/daemon_set.go b/test/e2e/daemon_set.go
index e2d5d928a08..87332921567 100644
--- a/test/e2e/daemon_set.go
+++ b/test/e2e/daemon_set.go
@@ -28,7 +28,6 @@ import (
 	"k8s.io/kubernetes/pkg/apimachinery/registered"
 	"k8s.io/kubernetes/pkg/apis/extensions"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
-	clientsetadapter "k8s.io/kubernetes/pkg/client/unversioned/adapters/internalclientset"
 	"k8s.io/kubernetes/pkg/kubectl"
 	"k8s.io/kubernetes/pkg/labels"
 	"k8s.io/kubernetes/pkg/runtime"
@@ -116,7 +115,7 @@ var _ = framework.KubeDescribe("Daemon set [Serial]", func() {
 		Expect(err).NotTo(HaveOccurred())
 		defer func() {
 			framework.Logf("Check that reaper kills all daemon pods for %s", dsName)
-			dsReaper, err := kubectl.ReaperFor(extensions.Kind("DaemonSet"), clientsetadapter.FromUnversionedClient(c))
+			dsReaper, err := kubectl.ReaperFor(extensions.Kind("DaemonSet"), f.ClientSet)
 			Expect(err).NotTo(HaveOccurred())
 			err = dsReaper.Stop(ns, dsName, 0, nil)
 			Expect(err).NotTo(HaveOccurred())
diff --git a/test/e2e/density.go b/test/e2e/density.go
index e6076299d2e..0c0296a1676 100644
--- a/test/e2e/density.go
+++ b/test/e2e/density.go
@@ -29,6 +29,7 @@ import (
 	"k8s.io/kubernetes/pkg/api/resource"
 	"k8s.io/kubernetes/pkg/api/unversioned"
 	"k8s.io/kubernetes/pkg/client/cache"
+	"k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
 	"k8s.io/kubernetes/pkg/fields"
 	"k8s.io/kubernetes/pkg/labels"
@@ -54,6 +55,7 @@ var MaxContainerFailures = 0
 type DensityTestConfig struct {
 	Configs      []framework.RCConfig
 	Client       *client.Client
+	ClientSet    internalclientset.Interface
 	Namespace    string
 	PollInterval time.Duration
 	PodCount     int
@@ -328,7 +330,7 @@ func cleanupDensityTest(dtc DensityTestConfig) {
 				framework.ExpectNoError(err)
 			} else {
 				By("Cleaning up the replication controller and pods")
-				err := framework.DeleteRCAndPods(dtc.Client, dtc.Namespace, rcName)
+				err := framework.DeleteRCAndPods(dtc.Client, dtc.ClientSet, dtc.Namespace, rcName)
 				framework.ExpectNoError(err)
 			}
 		}
@@ -487,7 +489,9 @@ var _ = framework.KubeDescribe("Density", func() {
 				}
 			}
 
-			dConfig := DensityTestConfig{Client: c,
+			dConfig := DensityTestConfig{
+				Client:       c,
+				ClientSet:    f.ClientSet,
 				Configs:      RCConfigs,
 				PodCount:     totalPods,
 				Namespace:    ns,
@@ -705,7 +709,9 @@ var _ = framework.KubeDescribe("Density", func() {
 				Silent:               true,
 			}
 		}
-		dConfig := DensityTestConfig{Client: c,
+		dConfig := DensityTestConfig{
+			Client:       c,
+			ClientSet:    f.ClientSet,
 			Configs:      RCConfigs,
 			PodCount:     totalPods,
 			Namespace:    ns,
diff --git a/test/e2e/deployment.go b/test/e2e/deployment.go
index 5b3e77a2d87..c6d980602ad 100644
--- a/test/e2e/deployment.go
+++ b/test/e2e/deployment.go
@@ -30,7 +30,6 @@ import (
 	"k8s.io/kubernetes/pkg/apis/extensions"
 	clientset "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
-	adapter "k8s.io/kubernetes/pkg/client/unversioned/adapters/internalclientset"
 	deploymentutil "k8s.io/kubernetes/pkg/controller/deployment/util"
 	"k8s.io/kubernetes/pkg/kubectl"
 	"k8s.io/kubernetes/pkg/labels"
@@ -160,7 +159,7 @@ func newDeploymentRollback(name string, annotations map[string]string, revision
 }
 
 // checkDeploymentRevision checks if the input deployment's and its new replica set's revision and images are as expected.
-func checkDeploymentRevision(c *clientset.Clientset, ns, deploymentName, revision, imageName, image string) (*extensions.Deployment, *extensions.ReplicaSet) {
+func checkDeploymentRevision(c clientset.Interface, ns, deploymentName, revision, imageName, image string) (*extensions.Deployment, *extensions.ReplicaSet) {
 	deployment, err := c.Extensions().Deployments(ns).Get(deploymentName)
 	Expect(err).NotTo(HaveOccurred())
 	// Check revision of the new replica set of this deployment
@@ -182,15 +181,15 @@ func checkDeploymentRevision(c *clientset.Clientset, ns, deploymentName, revisio
 	return deployment, newRS
 }
 
-func stopDeploymentOverlap(c *clientset.Clientset, oldC client.Interface, ns, deploymentName, overlapWith string) {
+func stopDeploymentOverlap(c clientset.Interface, oldC client.Interface, ns, deploymentName, overlapWith string) {
 	stopDeploymentMaybeOverlap(c, oldC, ns, deploymentName, overlapWith)
 }
 
-func stopDeployment(c *clientset.Clientset, oldC client.Interface, ns, deploymentName string) {
+func stopDeployment(c clientset.Interface, oldC client.Interface, ns, deploymentName string) {
 	stopDeploymentMaybeOverlap(c, oldC, ns, deploymentName, "")
 }
 
-func stopDeploymentMaybeOverlap(c *clientset.Clientset, oldC client.Interface, ns, deploymentName, overlapWith string) {
+func stopDeploymentMaybeOverlap(c clientset.Interface, oldC client.Interface, ns, deploymentName, overlapWith string) {
 	deployment, err := c.Extensions().Deployments(ns).Get(deploymentName)
 	Expect(err).NotTo(HaveOccurred())
 
@@ -254,7 +253,7 @@ func testNewDeployment(f *framework.Framework) {
 	ns := f.Namespace.Name
 	// TODO: remove unversionedClient when the refactoring is done. Currently some
 	// functions like verifyPod still expects a unversioned#Client.
-	c := adapter.FromUnversionedClient(f.Client)
+	c := f.ClientSet
 
 	deploymentName := "test-new-deployment"
 	podLabels := map[string]string{"name": nginxImageName}
@@ -289,7 +288,7 @@ func testRollingUpdateDeployment(f *framework.Framework) {
 	// TODO: remove unversionedClient when the refactoring is done. Currently some
 	// functions like verifyPod still expects a unversioned#Client.
 	unversionedClient := f.Client
-	c := adapter.FromUnversionedClient(unversionedClient)
+	c := f.ClientSet
 	// Create nginx pods.
 	deploymentPodLabels := map[string]string{"name": "sample-pod"}
 	rsPodLabels := map[string]string{
@@ -339,7 +338,7 @@ func testRollingUpdateDeploymentEvents(f *framework.Framework) {
 	// TODO: remove unversionedClient when the refactoring is done. Currently some
 	// functions like verifyPod still expects a unversioned#Client.
 	unversionedClient := f.Client
-	c := adapter.FromUnversionedClient(unversionedClient)
+	c := f.ClientSet
 	// Create nginx pods.
 	deploymentPodLabels := map[string]string{"name": "sample-pod-2"}
 	rsPodLabels := map[string]string{
@@ -401,7 +400,7 @@ func testRecreateDeployment(f *framework.Framework) {
 	// TODO: remove unversionedClient when the refactoring is done. Currently some
 	// functions like verifyPod still expects a unversioned#Client.
 	unversionedClient := f.Client
-	c := adapter.FromUnversionedClient(unversionedClient)
+	c := f.ClientSet
 	// Create nginx pods.
 	deploymentPodLabels := map[string]string{"name": "sample-pod-3"}
 	rsPodLabels := map[string]string{
@@ -456,7 +455,7 @@ func testRecreateDeployment(f *framework.Framework) {
 func testDeploymentCleanUpPolicy(f *framework.Framework) {
 	ns := f.Namespace.Name
 	unversionedClient := f.Client
-	c := adapter.FromUnversionedClient(unversionedClient)
+	c := f.ClientSet
 	// Create nginx pods.
 	deploymentPodLabels := map[string]string{"name": "cleanup-pod"}
 	rsPodLabels := map[string]string{
@@ -480,7 +479,7 @@ func testDeploymentCleanUpPolicy(f *framework.Framework) {
 	deploymentName := "test-cleanup-deployment"
 	framework.Logf("Creating deployment %s", deploymentName)
 
-	pods, err := c.Pods(ns).List(api.ListOptions{LabelSelector: labels.Everything()})
+	pods, err := c.Core().Pods(ns).List(api.ListOptions{LabelSelector: labels.Everything()})
 	if err != nil {
 		Expect(err).NotTo(HaveOccurred(), "Failed to query for pods: %v", err)
 	}
@@ -488,7 +487,7 @@ func testDeploymentCleanUpPolicy(f *framework.Framework) {
 		ResourceVersion: pods.ListMeta.ResourceVersion,
 	}
 	stopCh := make(chan struct{})
-	w, err := c.Pods(ns).Watch(options)
+	w, err := c.Core().Pods(ns).Watch(options)
 	go func() {
 		// There should be only one pod being created, which is the pod with the redis image.
 		// The old RS shouldn't create new pod when deployment controller adding pod template hash label to its selector.
@@ -531,7 +530,7 @@ func testRolloverDeployment(f *framework.Framework) {
 	// TODO: remove unversionedClient when the refactoring is done. Currently some
 	// functions like verifyPod still expects a unversioned#Client.
 	unversionedClient := f.Client
-	c := adapter.FromUnversionedClient(unversionedClient)
+	c := f.ClientSet
 	podName := "rollover-pod"
 	deploymentPodLabels := map[string]string{"name": podName}
 	rsPodLabels := map[string]string{
@@ -599,8 +598,7 @@ func testPausedDeployment(f *framework.Framework) {
 	ns := f.Namespace.Name
 	// TODO: remove unversionedClient when the refactoring is done. Currently some
 	// functions like verifyPod still expects a unversioned#Client.
-	unversionedClient := f.Client
-	c := adapter.FromUnversionedClient(unversionedClient)
+	c := f.ClientSet
 	deploymentName := "test-paused-deployment"
 	podLabels := map[string]string{"name": nginxImageName}
 	d := newDeployment(deploymentName, 1, podLabels, nginxImageName, nginxImage, extensions.RollingUpdateDeploymentStrategyType, nil)
@@ -697,8 +695,7 @@ func testPausedDeployment(f *framework.Framework) {
 // and then rollback to last revision.
 func testRollbackDeployment(f *framework.Framework) {
 	ns := f.Namespace.Name
-	unversionedClient := f.Client
-	c := adapter.FromUnversionedClient(unversionedClient)
+	c := f.ClientSet
 	podName := "nginx"
 	deploymentPodLabels := map[string]string{"name": podName}
 
@@ -806,7 +803,7 @@ func testRollbackDeployment(f *framework.Framework) {
 // TODO: When we finished reporting rollback status in deployment status, check the rollback status here in each case.
 func testRollbackDeploymentRSNoRevision(f *framework.Framework) {
 	ns := f.Namespace.Name
-	c := adapter.FromUnversionedClient(f.Client)
+	c := f.ClientSet
 	podName := "nginx"
 	deploymentPodLabels := map[string]string{"name": podName}
 	rsPodLabels := map[string]string{
@@ -943,7 +940,7 @@ func testDeploymentLabelAdopted(f *framework.Framework) {
 	// TODO: remove unversionedClient when the refactoring is done. Currently some
 	// functions like verifyPod still expects a unversioned#Client.
 	unversionedClient := f.Client
-	c := adapter.FromUnversionedClient(unversionedClient)
+	c := f.ClientSet
 	// Create nginx pods.
 	podName := "nginx"
 	podLabels := map[string]string{"name": podName}
@@ -998,7 +995,7 @@ func testDeploymentLabelAdopted(f *framework.Framework) {
 
 func testScalePausedDeployment(f *framework.Framework) {
 	ns := f.Namespace.Name
-	c := adapter.FromUnversionedClient(f.Client)
+	c := f.ClientSet
 
 	podLabels := map[string]string{"name": nginxImageName}
 	replicas := int32(3)
@@ -1049,7 +1046,7 @@ func testScalePausedDeployment(f *framework.Framework) {
 
 func testScaledRolloutDeployment(f *framework.Framework) {
 	ns := f.Namespace.Name
-	c := adapter.FromUnversionedClient(f.Client)
+	c := f.ClientSet
 
 	podLabels := map[string]string{"name": nginxImageName}
 	replicas := int32(10)
@@ -1216,7 +1213,7 @@ func testOverlappingDeployment(f *framework.Framework) {
 	ns := f.Namespace.Name
 	// TODO: remove unversionedClient when the refactoring is done. Currently some
 	// functions like verifyPod still expects a unversioned#Client.
-	c := adapter.FromUnversionedClient(f.Client)
+	c := f.ClientSet
 
 	deploymentName := "first-deployment"
 	podLabels := map[string]string{"name": redisImageName}
diff --git a/test/e2e/empty_dir_wrapper.go b/test/e2e/empty_dir_wrapper.go
index 342ae7d7ea8..373ffb42285 100644
--- a/test/e2e/empty_dir_wrapper.go
+++ b/test/e2e/empty_dir_wrapper.go
@@ -377,7 +377,7 @@ func testNoWrappedVolumeRace(f *framework.Framework, volumes []api.Volume, volum
 	Expect(err).NotTo(HaveOccurred(), "error creating replication controller")
 
 	defer func() {
-		err := framework.DeleteRCAndPods(f.Client, f.Namespace.Name, rcName)
+		err := framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, rcName)
 		framework.ExpectNoError(err)
 	}()
 
diff --git a/test/e2e/examples.go b/test/e2e/examples.go
index 086fb9339ff..a5f9a644aa7 100644
--- a/test/e2e/examples.go
+++ b/test/e2e/examples.go
@@ -503,7 +503,7 @@ var _ = framework.KubeDescribe("[Feature:Example]", func() {
 			Expect(err).NotTo(HaveOccurred())
 
 			By("scaling rethinkdb")
-			framework.ScaleRC(c, ns, "rethinkdb-rc", 2, true)
+			framework.ScaleRC(c, f.ClientSet, ns, "rethinkdb-rc", 2, true)
 			checkDbInstances()
 
 			By("starting admin")
@@ -546,7 +546,7 @@ var _ = framework.KubeDescribe("[Feature:Example]", func() {
 			Expect(err).NotTo(HaveOccurred())
 
 			By("scaling hazelcast")
-			framework.ScaleRC(c, ns, "hazelcast", 2, true)
+			framework.ScaleRC(c, f.ClientSet, ns, "hazelcast", 2, true)
 			forEachPod("name", "hazelcast", func(pod api.Pod) {
 				_, err := framework.LookForStringInLog(ns, pod.Name, "hazelcast", "Members [2]", serverStartTimeout)
 				Expect(err).NotTo(HaveOccurred())
diff --git a/test/e2e/framework/framework.go b/test/e2e/framework/framework.go
index fd62e048ccf..51e94950c30 100644
--- a/test/e2e/framework/framework.go
+++ b/test/e2e/framework/framework.go
@@ -35,6 +35,7 @@ import (
 	"k8s.io/kubernetes/pkg/api/unversioned"
 	"k8s.io/kubernetes/pkg/api/v1"
 	"k8s.io/kubernetes/pkg/apimachinery/registered"
+	"k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
 	"k8s.io/kubernetes/pkg/client/clientset_generated/release_1_5"
 	"k8s.io/kubernetes/pkg/client/restclient"
 	"k8s.io/kubernetes/pkg/client/typed/dynamic"
@@ -59,8 +60,13 @@ const (
 type Framework struct {
 	BaseName string
 
-	Client        *client.Client
-	Clientset_1_5 *release_1_5.Clientset
+	// Client is manually created and should not be used unless absolutely necessary. Use ClientSet_1_5
+	// where possible.
+	Client *client.Client
+	// ClientSet uses internal objects, you should use ClientSet_1_5 where possible.
+	ClientSet internalclientset.Interface
+
+	ClientSet_1_5 *release_1_5.Clientset
 	StagingClient *staging.Clientset
 	ClientPool    dynamic.ClientPool
 
@@ -193,7 +199,9 @@ func (f *Framework) BeforeEach() {
 		c, err := loadClientFromConfig(config)
 		Expect(err).NotTo(HaveOccurred())
 		f.Client = c
-		f.Clientset_1_5, err = release_1_5.NewForConfig(config)
+		f.ClientSet, err = internalclientset.NewForConfig(config)
+		Expect(err).NotTo(HaveOccurred())
+		f.ClientSet_1_5, err = release_1_5.NewForConfig(config)
 		Expect(err).NotTo(HaveOccurred())
 		clientRepoConfig := getClientRepoConfig(config)
 		f.StagingClient, err = staging.NewForConfig(clientRepoConfig)
@@ -358,7 +366,7 @@ func (f *Framework) AfterEach() {
 	// Print events if the test failed.
 	if CurrentGinkgoTestDescription().Failed && TestContext.DumpLogsOnFailure {
 		// Pass both unversioned client and and versioned clientset, till we have removed all uses of the unversioned client.
-		DumpAllNamespaceInfo(f.Client, f.Clientset_1_5, f.Namespace.Name)
+		DumpAllNamespaceInfo(f.Client, f.ClientSet_1_5, f.Namespace.Name)
 		By(fmt.Sprintf("Dumping a list of prepulled images on each node"))
 		LogContainersInPodsWithLabels(f.Client, api.NamespaceSystem, ImagePullerLabels, "image-puller")
 		if f.federated {
diff --git a/test/e2e/framework/util.go b/test/e2e/framework/util.go
index c12c8375fdd..6c0f9891ba1 100644
--- a/test/e2e/framework/util.go
+++ b/test/e2e/framework/util.go
@@ -57,7 +57,6 @@ import (
 	"k8s.io/kubernetes/pkg/client/typed/discovery"
 	"k8s.io/kubernetes/pkg/client/typed/dynamic"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
-	clientsetadapter "k8s.io/kubernetes/pkg/client/unversioned/adapters/internalclientset"
 	"k8s.io/kubernetes/pkg/client/unversioned/clientcmd"
 	clientcmdapi "k8s.io/kubernetes/pkg/client/unversioned/clientcmd/api"
 	gcecloud "k8s.io/kubernetes/pkg/cloudprovider/providers/gce"
@@ -3224,9 +3223,9 @@ func RemoveTaintOffNode(c *client.Client, nodeName string, taint api.Taint) {
 	}
 }
 
-func ScaleRC(c *client.Client, ns, name string, size uint, wait bool) error {
+func ScaleRC(c *client.Client, clientset clientset.Interface, ns, name string, size uint, wait bool) error {
 	By(fmt.Sprintf("Scaling replication controller %s in namespace %s to %d", name, ns, size))
-	scaler, err := kubectl.ScalerFor(api.Kind("ReplicationController"), clientsetadapter.FromUnversionedClient(c))
+	scaler, err := kubectl.ScalerFor(api.Kind("ReplicationController"), clientset)
 	if err != nil {
 		return err
 	}
@@ -3333,7 +3332,7 @@ func WaitForPodsWithLabel(c *client.Client, ns string, label labels.Selector) (p
 }
 
 // DeleteRCAndPods a Replication Controller and all pods it spawned
-func DeleteRCAndPods(c *client.Client, ns, name string) error {
+func DeleteRCAndPods(c *client.Client, clientset clientset.Interface, ns, name string) error {
 	By(fmt.Sprintf("deleting replication controller %s in namespace %s", name, ns))
 	rc, err := c.ReplicationControllers(ns).Get(name)
 	if err != nil {
@@ -3343,7 +3342,7 @@ func DeleteRCAndPods(c *client.Client, ns, name string) error {
 		}
 		return err
 	}
-	reaper, err := kubectl.ReaperForReplicationController(clientsetadapter.FromUnversionedClient(c).Core(), 10*time.Minute)
+	reaper, err := kubectl.ReaperForReplicationController(clientset.Core(), 10*time.Minute)
 	if err != nil {
 		if apierrs.IsNotFound(err) {
 			Logf("RC %s was already deleted: %v", name, err)
@@ -3481,7 +3480,7 @@ func waitForPodsGone(ps *PodStore, interval, timeout time.Duration) error {
 }
 
 // Delete a ReplicaSet and all pods it spawned
-func DeleteReplicaSet(c *client.Client, ns, name string) error {
+func DeleteReplicaSet(c *client.Client, clientset clientset.Interface, ns, name string) error {
 	By(fmt.Sprintf("deleting ReplicaSet %s in namespace %s", name, ns))
 	rc, err := c.Extensions().ReplicaSets(ns).Get(name)
 	if err != nil {
@@ -3491,7 +3490,7 @@ func DeleteReplicaSet(c *client.Client, ns, name string) error {
 		}
 		return err
 	}
-	reaper, err := kubectl.ReaperFor(extensions.Kind("ReplicaSet"), clientsetadapter.FromUnversionedClient(c))
+	reaper, err := kubectl.ReaperFor(extensions.Kind("ReplicaSet"), clientset)
 	if err != nil {
 		if apierrs.IsNotFound(err) {
 			Logf("ReplicaSet %s was already deleted: %v", name, err)
@@ -3770,11 +3769,11 @@ func CheckNewRSAnnotations(c clientset.Interface, ns, deploymentName string, exp
 	return nil
 }
 
-func WaitForPodsReady(c *clientset.Clientset, ns, name string, minReadySeconds int) error {
+func WaitForPodsReady(c clientset.Interface, ns, name string, minReadySeconds int) error {
 	label := labels.SelectorFromSet(labels.Set(map[string]string{"name": name}))
 	options := api.ListOptions{LabelSelector: label}
 	return wait.Poll(Poll, 5*time.Minute, func() (bool, error) {
-		pods, err := c.Pods(ns).List(options)
+		pods, err := c.Core().Pods(ns).List(options)
 		if err != nil {
 			return false, nil
 		}
@@ -3788,7 +3787,7 @@ func WaitForPodsReady(c *clientset.Clientset, ns, name string, minReadySeconds i
 }
 
 // Waits for the deployment to clean up old rcs.
-func WaitForDeploymentOldRSsNum(c *clientset.Clientset, ns, deploymentName string, desiredRSNum int) error {
+func WaitForDeploymentOldRSsNum(c clientset.Interface, ns, deploymentName string, desiredRSNum int) error {
 	return wait.Poll(Poll, 5*time.Minute, func() (bool, error) {
 		deployment, err := c.Extensions().Deployments(ns).Get(deploymentName)
 		if err != nil {
@@ -3814,7 +3813,7 @@ func logReplicaSetsOfDeployment(deployment *extensions.Deployment, allOldRSs []*
 	}
 }
 
-func WaitForObservedDeployment(c *clientset.Clientset, ns, deploymentName string, desiredGeneration int64) error {
+func WaitForObservedDeployment(c clientset.Interface, ns, deploymentName string, desiredGeneration int64) error {
 	return deploymentutil.WaitForObservedDeployment(func() (*extensions.Deployment, error) { return c.Extensions().Deployments(ns).Get(deploymentName) }, desiredGeneration, Poll, 1*time.Minute)
 }
 
@@ -3875,7 +3874,7 @@ func WaitForPartialEvents(c *client.Client, ns string, objOrRef runtime.Object,
 
 type updateDeploymentFunc func(d *extensions.Deployment)
 
-func UpdateDeploymentWithRetries(c *clientset.Clientset, namespace, name string, applyUpdate updateDeploymentFunc) (deployment *extensions.Deployment, err error) {
+func UpdateDeploymentWithRetries(c clientset.Interface, namespace, name string, applyUpdate updateDeploymentFunc) (deployment *extensions.Deployment, err error) {
 	deployments := c.Extensions().Deployments(namespace)
 	err = wait.Poll(10*time.Millisecond, 1*time.Minute, func() (bool, error) {
 		if deployment, err = deployments.Get(name); err != nil {
@@ -4722,7 +4721,7 @@ func GetNodePortURL(client *client.Client, ns, name string, svcPort int) (string
 
 // ScaleRCByLabels scales an RC via ns/label lookup. If replicas == 0 it waits till
 // none are running, otherwise it does what a synchronous scale operation would do.
-func ScaleRCByLabels(client *client.Client, ns string, l map[string]string, replicas uint) error {
+func ScaleRCByLabels(client *client.Client, clientset clientset.Interface, ns string, l map[string]string, replicas uint) error {
 	listOpts := api.ListOptions{LabelSelector: labels.SelectorFromSet(labels.Set(l))}
 	rcs, err := client.ReplicationControllers(ns).List(listOpts)
 	if err != nil {
@@ -4734,7 +4733,7 @@ func ScaleRCByLabels(client *client.Client, ns string, l map[string]string, repl
 	Logf("Scaling %v RCs with labels %v in ns %v to %v replicas.", len(rcs.Items), l, ns, replicas)
 	for _, labelRC := range rcs.Items {
 		name := labelRC.Name
-		if err := ScaleRC(client, ns, name, replicas, false); err != nil {
+		if err := ScaleRC(client, clientset, ns, name, replicas, false); err != nil {
 			return err
 		}
 		rc, err := client.ReplicationControllers(ns).Get(name)
diff --git a/test/e2e/garbage_collector.go b/test/e2e/garbage_collector.go
index edc33eca072..77f9347f97a 100644
--- a/test/e2e/garbage_collector.go
+++ b/test/e2e/garbage_collector.go
@@ -118,7 +118,7 @@ func gatherMetrics(f *framework.Framework) {
 var _ = framework.KubeDescribe("Garbage collector", func() {
 	f := framework.NewDefaultFramework("gc")
 	It("[Feature:GarbageCollector] should delete pods created by rc when not orphaning", func() {
-		clientSet := f.Clientset_1_5
+		clientSet := f.ClientSet_1_5
 		rcClient := clientSet.Core().ReplicationControllers(f.Namespace.Name)
 		podClient := clientSet.Core().Pods(f.Namespace.Name)
 		rcName := "simpletest.rc"
@@ -169,7 +169,7 @@ var _ = framework.KubeDescribe("Garbage collector", func() {
 	})
 
 	It("[Feature:GarbageCollector] should orphan pods created by rc if delete options say so", func() {
-		clientSet := f.Clientset_1_5
+		clientSet := f.ClientSet_1_5
 		rcClient := clientSet.Core().ReplicationControllers(f.Namespace.Name)
 		podClient := clientSet.Core().Pods(f.Namespace.Name)
 		rcName := "simpletest.rc"
@@ -231,7 +231,7 @@ var _ = framework.KubeDescribe("Garbage collector", func() {
 	})
 
 	It("[Feature:GarbageCollector] should orphan pods created by rc if deleteOptions.OrphanDependents is nil", func() {
-		clientSet := f.Clientset_1_5
+		clientSet := f.ClientSet_1_5
 		rcClient := clientSet.Core().ReplicationControllers(f.Namespace.Name)
 		podClient := clientSet.Core().Pods(f.Namespace.Name)
 		rcName := "simpletest.rc"
diff --git a/test/e2e/generated_clientset.go b/test/e2e/generated_clientset.go
index 0e094684159..10e127de9e7 100644
--- a/test/e2e/generated_clientset.go
+++ b/test/e2e/generated_clientset.go
@@ -122,7 +122,7 @@ func observeObjectDeletion(w watch.Interface) (obj runtime.Object) {
 var _ = framework.KubeDescribe("Generated release_1_5 clientset", func() {
 	f := framework.NewDefaultFramework("clientset")
 	It("should create pods, delete pods, watch pods", func() {
-		podClient := f.Clientset_1_5.Core().Pods(f.Namespace.Name)
+		podClient := f.ClientSet_1_5.Core().Pods(f.Namespace.Name)
 		By("constructing the pod")
 		name := "pod" + string(uuid.NewUUID())
 		value := strconv.Itoa(time.Now().Nanosecond())
diff --git a/test/e2e/job.go b/test/e2e/job.go
index 72aa9a7602e..0e50d80229d 100644
--- a/test/e2e/job.go
+++ b/test/e2e/job.go
@@ -23,7 +23,6 @@ import (
 	"k8s.io/kubernetes/pkg/api/errors"
 	"k8s.io/kubernetes/pkg/apis/batch"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
-	clientsetadapter "k8s.io/kubernetes/pkg/client/unversioned/adapters/internalclientset"
 	"k8s.io/kubernetes/pkg/kubectl"
 	"k8s.io/kubernetes/pkg/labels"
 	"k8s.io/kubernetes/pkg/util/wait"
@@ -125,7 +124,7 @@ var _ = framework.KubeDescribe("Job", func() {
 		Expect(err).NotTo(HaveOccurred())
 
 		By("scale job up")
-		scaler, err := kubectl.ScalerFor(batch.Kind("Job"), clientsetadapter.FromUnversionedClient(f.Client))
+		scaler, err := kubectl.ScalerFor(batch.Kind("Job"), f.ClientSet)
 		Expect(err).NotTo(HaveOccurred())
 		waitForScale := kubectl.NewRetryParams(5*time.Second, 1*time.Minute)
 		waitForReplicas := kubectl.NewRetryParams(5*time.Second, 5*time.Minute)
@@ -150,7 +149,7 @@ var _ = framework.KubeDescribe("Job", func() {
 		Expect(err).NotTo(HaveOccurred())
 
 		By("scale job down")
-		scaler, err := kubectl.ScalerFor(batch.Kind("Job"), clientsetadapter.FromUnversionedClient(f.Client))
+		scaler, err := kubectl.ScalerFor(batch.Kind("Job"), f.ClientSet)
 		Expect(err).NotTo(HaveOccurred())
 		waitForScale := kubectl.NewRetryParams(5*time.Second, 1*time.Minute)
 		waitForReplicas := kubectl.NewRetryParams(5*time.Second, 5*time.Minute)
@@ -173,7 +172,7 @@ var _ = framework.KubeDescribe("Job", func() {
 		Expect(err).NotTo(HaveOccurred())
 
 		By("delete a job")
-		reaper, err := kubectl.ReaperFor(batch.Kind("Job"), clientsetadapter.FromUnversionedClient(f.Client))
+		reaper, err := kubectl.ReaperFor(batch.Kind("Job"), f.ClientSet)
 		Expect(err).NotTo(HaveOccurred())
 		timeout := 1 * time.Minute
 		err = reaper.Stop(f.Namespace.Name, job.Name, timeout, api.NewDeleteOptions(0))
diff --git a/test/e2e/kubectl.go b/test/e2e/kubectl.go
index cef8568275a..a35931f78b2 100644
--- a/test/e2e/kubectl.go
+++ b/test/e2e/kubectl.go
@@ -47,7 +47,6 @@ import (
 	"k8s.io/kubernetes/pkg/api/resource"
 	"k8s.io/kubernetes/pkg/api/unversioned"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
-	clientsetadapter "k8s.io/kubernetes/pkg/client/unversioned/adapters/internalclientset"
 	"k8s.io/kubernetes/pkg/controller"
 	"k8s.io/kubernetes/pkg/kubectl/cmd/util"
 	"k8s.io/kubernetes/pkg/labels"
@@ -199,7 +198,7 @@ var _ = framework.KubeDescribe("Kubectl client", func() {
 		pods, err := clusterState().WaitFor(atLeast, framework.PodStartTimeout)
 		if err != nil || len(pods) < atLeast {
 			// TODO: Generalize integrating debug info into these tests so we always get debug info when we need it
-			framework.DumpAllNamespaceInfo(c, f.Clientset_1_5, ns)
+			framework.DumpAllNamespaceInfo(c, f.ClientSet_1_5, ns)
 			framework.Failf("Verified %v of %v pods , error : %v", len(pods), atLeast, err)
 		}
 	}
@@ -429,8 +428,8 @@ var _ = framework.KubeDescribe("Kubectl client", func() {
 				WithStdinData("abcd1234\n").
 				ExecOrDie()
 			Expect(runOutput).ToNot(ContainSubstring("stdin closed"))
-			f := func(pods []*api.Pod) sort.Interface { return sort.Reverse(controller.ActivePods(pods)) }
-			runTestPod, _, err := util.GetFirstPod(clientsetadapter.FromUnversionedClient(c), ns, labels.SelectorFromSet(map[string]string{"run": "run-test-3"}), 1*time.Minute, f)
+			g := func(pods []*api.Pod) sort.Interface { return sort.Reverse(controller.ActivePods(pods)) }
+			runTestPod, _, err := util.GetFirstPod(f.ClientSet.Core(), ns, labels.SelectorFromSet(map[string]string{"run": "run-test-3"}), 1*time.Minute, g)
 			if err != nil {
 				os.Exit(1)
 			}
diff --git a/test/e2e/kubelet.go b/test/e2e/kubelet.go
index 9421ccafd1c..dc9ee7cb0ab 100644
--- a/test/e2e/kubelet.go
+++ b/test/e2e/kubelet.go
@@ -205,7 +205,7 @@ var _ = framework.KubeDescribe("kubelet", func() {
 				}
 
 				By("Deleting the RC")
-				framework.DeleteRCAndPods(f.Client, f.Namespace.Name, rcName)
+				framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, rcName)
 				// Check that the pods really are gone by querying /runningpods on the
 				// node. The /runningpods handler checks the container runtime (or its
 				// cache) and  returns a list of running pods. Some possible causes of
diff --git a/test/e2e/kubelet_perf.go b/test/e2e/kubelet_perf.go
index 8446a78591b..d7cb3cf340a 100644
--- a/test/e2e/kubelet_perf.go
+++ b/test/e2e/kubelet_perf.go
@@ -115,7 +115,7 @@ func runResourceTrackingTest(f *framework.Framework, podsPerNode int, nodeNames
 	verifyCPULimits(expectedCPU, cpuSummary)
 
 	By("Deleting the RC")
-	framework.DeleteRCAndPods(f.Client, f.Namespace.Name, rcName)
+	framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, rcName)
 }
 
 func verifyMemoryLimits(c *client.Client, expected framework.ResourceUsagePerContainer, actual framework.ResourceUsagePerNode) {
diff --git a/test/e2e/load.go b/test/e2e/load.go
index 7d68021390e..315d6bf82b4 100644
--- a/test/e2e/load.go
+++ b/test/e2e/load.go
@@ -26,6 +26,8 @@ import (
 	"time"
 
 	"k8s.io/kubernetes/pkg/api"
+	"k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
+	unversionedcore "k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset/typed/core/unversioned"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
 	"k8s.io/kubernetes/pkg/labels"
 	"k8s.io/kubernetes/pkg/util/intstr"
@@ -321,7 +323,7 @@ func scaleRC(wg *sync.WaitGroup, config *framework.RCConfig, scalingTime time.Du
 
 	sleepUpTo(scalingTime)
 	newSize := uint(rand.Intn(config.Replicas) + config.Replicas/2)
-	framework.ExpectNoError(framework.ScaleRC(config.Client, config.Namespace, config.Name, newSize, true),
+	framework.ExpectNoError(framework.ScaleRC(config.Client, coreClientSetFromUnversioned(config.Client), config.Namespace, config.Name, newSize, true),
 		fmt.Sprintf("scaling rc %s for the first time", config.Name))
 	selector := labels.SelectorFromSet(labels.Set(map[string]string{"name": config.Name}))
 	options := api.ListOptions{
@@ -349,6 +351,17 @@ func deleteRC(wg *sync.WaitGroup, config *framework.RCConfig, deletingTime time.
 	if framework.TestContext.GarbageCollectorEnabled {
 		framework.ExpectNoError(framework.DeleteRCAndWaitForGC(config.Client, config.Namespace, config.Name), fmt.Sprintf("deleting rc %s", config.Name))
 	} else {
-		framework.ExpectNoError(framework.DeleteRCAndPods(config.Client, config.Namespace, config.Name), fmt.Sprintf("deleting rc %s", config.Name))
+		framework.ExpectNoError(framework.DeleteRCAndPods(config.Client, coreClientSetFromUnversioned(config.Client), config.Namespace, config.Name), fmt.Sprintf("deleting rc %s", config.Name))
 	}
 }
+
+// coreClientSetFromUnversioned adapts just enough of a a unversioned.Client to work with the scale RC function
+func coreClientSetFromUnversioned(c *client.Client) internalclientset.Interface {
+	var clientset internalclientset.Clientset
+	if c != nil {
+		clientset.CoreClient = unversionedcore.New(c.RESTClient)
+	} else {
+		clientset.CoreClient = unversionedcore.New(nil)
+	}
+	return &clientset
+}
diff --git a/test/e2e/proxy.go b/test/e2e/proxy.go
index 42a3d9fe284..05af7e096e7 100644
--- a/test/e2e/proxy.go
+++ b/test/e2e/proxy.go
@@ -154,7 +154,7 @@ func proxyContext(version string) {
 			CreatedPods: &pods,
 		}
 		Expect(framework.RunRC(cfg)).NotTo(HaveOccurred())
-		defer framework.DeleteRCAndPods(f.Client, f.Namespace.Name, cfg.Name)
+		defer framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, cfg.Name)
 
 		Expect(f.WaitForAnEndpoint(service.Name)).NotTo(HaveOccurred())
 
diff --git a/test/e2e/rc.go b/test/e2e/rc.go
index eac8d71d87e..e9e4187f91b 100644
--- a/test/e2e/rc.go
+++ b/test/e2e/rc.go
@@ -86,7 +86,7 @@ func ServeImageOrFail(f *framework.Framework, test string, image string) {
 	// Cleanup the replication controller when we are done.
 	defer func() {
 		// Resize the replication controller to zero to get rid of pods.
-		if err := framework.DeleteRCAndPods(f.Client, f.Namespace.Name, controller.Name); err != nil {
+		if err := framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, controller.Name); err != nil {
 			framework.Logf("Failed to cleanup replication controller %v: %v.", controller.Name, err)
 		}
 	}()
diff --git a/test/e2e/replica_set.go b/test/e2e/replica_set.go
index 33f35884ff4..defd7dd2d4f 100644
--- a/test/e2e/replica_set.go
+++ b/test/e2e/replica_set.go
@@ -86,7 +86,7 @@ func ReplicaSetServeImageOrFail(f *framework.Framework, test string, image strin
 	// Cleanup the ReplicaSet when we are done.
 	defer func() {
 		// Resize the ReplicaSet to zero to get rid of pods.
-		if err := framework.DeleteReplicaSet(f.Client, f.Namespace.Name, rs.Name); err != nil {
+		if err := framework.DeleteReplicaSet(f.Client, f.ClientSet, f.Namespace.Name, rs.Name); err != nil {
 			framework.Logf("Failed to cleanup ReplicaSet %v: %v.", rs.Name, err)
 		}
 	}()
diff --git a/test/e2e/rescheduler.go b/test/e2e/rescheduler.go
index f9c0050d532..5ea934a21a1 100644
--- a/test/e2e/rescheduler.go
+++ b/test/e2e/rescheduler.go
@@ -48,7 +48,7 @@ var _ = framework.KubeDescribe("Rescheduler [Serial]", func() {
 	It("should ensure that critical pod is scheduled in case there is no resources available", func() {
 		By("reserving all available cpu")
 		err := reserveAllCpu(f, "reserve-all-cpu", totalMillicores)
-		defer framework.DeleteRCAndPods(f.Client, ns, "reserve-all-cpu")
+		defer framework.DeleteRCAndPods(f.Client, f.ClientSet, ns, "reserve-all-cpu")
 		framework.ExpectNoError(err)
 
 		By("creating a new instance of DNS and waiting for DNS to be scheduled")
@@ -61,8 +61,8 @@ var _ = framework.KubeDescribe("Rescheduler [Serial]", func() {
 		rc := rcs.Items[0]
 		replicas := uint(rc.Spec.Replicas)
 
-		err = framework.ScaleRC(f.Client, api.NamespaceSystem, rc.Name, replicas+1, true)
-		defer framework.ExpectNoError(framework.ScaleRC(f.Client, api.NamespaceSystem, rc.Name, replicas, true))
+		err = framework.ScaleRC(f.Client, f.ClientSet, api.NamespaceSystem, rc.Name, replicas+1, true)
+		defer framework.ExpectNoError(framework.ScaleRC(f.Client, f.ClientSet, api.NamespaceSystem, rc.Name, replicas, true))
 		framework.ExpectNoError(err)
 	})
 })
@@ -72,7 +72,7 @@ func reserveAllCpu(f *framework.Framework, id string, millicores int) error {
 	replicas := millicores / 100
 
 	ReserveCpu(f, id, 1, 100)
-	framework.ExpectNoError(framework.ScaleRC(f.Client, f.Namespace.Name, id, uint(replicas), false))
+	framework.ExpectNoError(framework.ScaleRC(f.Client, f.ClientSet, f.Namespace.Name, id, uint(replicas), false))
 
 	for start := time.Now(); time.Since(start) < timeout; time.Sleep(10 * time.Second) {
 		pods, err := framework.GetPodsInNamespace(f.Client, f.Namespace.Name, framework.ImagePullerLabels)
diff --git a/test/e2e/scheduler_predicates.go b/test/e2e/scheduler_predicates.go
index 608ef691c18..92836ab5d5a 100644
--- a/test/e2e/scheduler_predicates.go
+++ b/test/e2e/scheduler_predicates.go
@@ -186,19 +186,18 @@ var _ = framework.KubeDescribe("SchedulerPredicates [Serial]", func() {
 	var totalPodCapacity int64
 	var RCName string
 	var ns string
+	f := framework.NewDefaultFramework("sched-pred")
 	ignoreLabels := framework.ImagePullerLabels
 
 	AfterEach(func() {
 		rc, err := c.ReplicationControllers(ns).Get(RCName)
 		if err == nil && rc.Spec.Replicas != 0 {
 			By("Cleaning up the replication controller")
-			err := framework.DeleteRCAndPods(c, ns, RCName)
+			err := framework.DeleteRCAndPods(c, f.ClientSet, ns, RCName)
 			framework.ExpectNoError(err)
 		}
 	})
 
-	f := framework.NewDefaultFramework("sched-pred")
-
 	BeforeEach(func() {
 		c = f.Client
 		ns = f.Namespace.Name
@@ -957,7 +956,7 @@ var _ = framework.KubeDescribe("SchedulerPredicates [Serial]", func() {
 		// cannot be scheduled onto it.
 		By("Launching two pods on two distinct nodes to get two node names")
 		CreateHostPortPods(f, "host-port", 2, true)
-		defer framework.DeleteRCAndPods(f.Client, f.Namespace.Name, "host-port")
+		defer framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, "host-port")
 		podList, err := c.Pods(ns).List(api.ListOptions{})
 		ExpectNoError(err)
 		Expect(len(podList.Items)).To(Equal(2))
diff --git a/test/e2e/service.go b/test/e2e/service.go
index a5182d12a34..95967817a46 100644
--- a/test/e2e/service.go
+++ b/test/e2e/service.go
@@ -33,6 +33,7 @@ import (
 	"k8s.io/kubernetes/pkg/api"
 	"k8s.io/kubernetes/pkg/api/errors"
 	"k8s.io/kubernetes/pkg/api/service"
+	"k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
 	"k8s.io/kubernetes/pkg/controller/endpoint"
 	"k8s.io/kubernetes/pkg/labels"
@@ -309,7 +310,7 @@ var _ = framework.KubeDescribe("Services", func() {
 
 		// Stop service 1 and make sure it is gone.
 		By("stopping service1")
-		framework.ExpectNoError(stopServeHostnameService(c, ns, "service1"))
+		framework.ExpectNoError(stopServeHostnameService(c, f.ClientSet, ns, "service1"))
 
 		By("verifying service1 is not up")
 		framework.ExpectNoError(verifyServeHostnameServiceDown(c, host, svc1IP, servicePort))
@@ -342,11 +343,11 @@ var _ = framework.KubeDescribe("Services", func() {
 		svc1 := "service1"
 		svc2 := "service2"
 
-		defer func() { framework.ExpectNoError(stopServeHostnameService(c, ns, svc1)) }()
+		defer func() { framework.ExpectNoError(stopServeHostnameService(c, f.ClientSet, ns, svc1)) }()
 		podNames1, svc1IP, err := startServeHostnameService(c, ns, svc1, servicePort, numPods)
 		Expect(err).NotTo(HaveOccurred())
 
-		defer func() { framework.ExpectNoError(stopServeHostnameService(c, ns, svc2)) }()
+		defer func() { framework.ExpectNoError(stopServeHostnameService(c, f.ClientSet, ns, svc2)) }()
 		podNames2, svc2IP, err := startServeHostnameService(c, ns, svc2, servicePort, numPods)
 		Expect(err).NotTo(HaveOccurred())
 
@@ -391,7 +392,7 @@ var _ = framework.KubeDescribe("Services", func() {
 		ns := f.Namespace.Name
 		numPods, servicePort := 3, 80
 
-		defer func() { framework.ExpectNoError(stopServeHostnameService(c, ns, "service1")) }()
+		defer func() { framework.ExpectNoError(stopServeHostnameService(c, f.ClientSet, ns, "service1")) }()
 		podNames1, svc1IP, err := startServeHostnameService(c, ns, "service1", servicePort, numPods)
 		Expect(err).NotTo(HaveOccurred())
 
@@ -416,7 +417,7 @@ var _ = framework.KubeDescribe("Services", func() {
 		framework.ExpectNoError(verifyServeHostnameServiceUp(c, ns, host, podNames1, svc1IP, servicePort))
 
 		// Create a new service and check if it's not reusing IP.
-		defer func() { framework.ExpectNoError(stopServeHostnameService(c, ns, "service2")) }()
+		defer func() { framework.ExpectNoError(stopServeHostnameService(c, f.ClientSet, ns, "service2")) }()
 		podNames2, svc2IP, err := startServeHostnameService(c, ns, "service2", servicePort, numPods)
 		Expect(err).NotTo(HaveOccurred())
 
@@ -1647,8 +1648,8 @@ func startServeHostnameService(c *client.Client, ns, name string, port, replicas
 	return podNames, serviceIP, nil
 }
 
-func stopServeHostnameService(c *client.Client, ns, name string) error {
-	if err := framework.DeleteRCAndPods(c, ns, name); err != nil {
+func stopServeHostnameService(c *client.Client, clientset internalclientset.Interface, ns, name string) error {
+	if err := framework.DeleteRCAndPods(c, clientset, ns, name); err != nil {
 		return err
 	}
 	if err := c.Services(ns).Delete(name); err != nil {
diff --git a/test/e2e/service_latency.go b/test/e2e/service_latency.go
index a71c2a3eef9..2203fda9d56 100644
--- a/test/e2e/service_latency.go
+++ b/test/e2e/service_latency.go
@@ -126,7 +126,7 @@ func runServiceLatencies(f *framework.Framework, inParallel, total int) (output
 	if err := framework.RunRC(cfg); err != nil {
 		return nil, err
 	}
-	defer framework.DeleteRCAndPods(f.Client, f.Namespace.Name, cfg.Name)
+	defer framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, cfg.Name)
 
 	// Run a single watcher, to reduce the number of API calls we have to
 	// make; this is to minimize the timing error. It's how kube-proxy
diff --git a/test/e2e/ubernetes_lite.go b/test/e2e/ubernetes_lite.go
index cd231104818..db4ce08660b 100644
--- a/test/e2e/ubernetes_lite.go
+++ b/test/e2e/ubernetes_lite.go
@@ -219,7 +219,7 @@ func SpreadRCOrFail(f *framework.Framework, replicaCount int32, image string) {
 	// Cleanup the replication controller when we are done.
 	defer func() {
 		// Resize the replication controller to zero to get rid of pods.
-		if err := framework.DeleteRCAndPods(f.Client, f.Namespace.Name, controller.Name); err != nil {
+		if err := framework.DeleteRCAndPods(f.Client, f.ClientSet, f.Namespace.Name, controller.Name); err != nil {
 			framework.Logf("Failed to cleanup replication controller %v: %v.", controller.Name, err)
 		}
 	}()
diff --git a/test/integration/framework/master_utils.go b/test/integration/framework/master_utils.go
index 667f1c7eb20..29422ff5169 100644
--- a/test/integration/framework/master_utils.go
+++ b/test/integration/framework/master_utils.go
@@ -42,10 +42,8 @@ import (
 	"k8s.io/kubernetes/pkg/client/record"
 	"k8s.io/kubernetes/pkg/client/restclient"
 	client "k8s.io/kubernetes/pkg/client/unversioned"
-	clientsetadapter "k8s.io/kubernetes/pkg/client/unversioned/adapters/internalclientset"
 	"k8s.io/kubernetes/pkg/controller"
 	replicationcontroller "k8s.io/kubernetes/pkg/controller/replication"
-	"k8s.io/kubernetes/pkg/fields"
 	"k8s.io/kubernetes/pkg/genericapiserver"
 	"k8s.io/kubernetes/pkg/genericapiserver/authorizer"
 	"k8s.io/kubernetes/pkg/kubectl"
@@ -300,8 +298,8 @@ func RCFromManifest(fileName string) *api.ReplicationController {
 }
 
 // StopRC stops the rc via kubectl's stop library
-func StopRC(rc *api.ReplicationController, restClient *client.Client) error {
-	reaper, err := kubectl.ReaperFor(api.Kind("ReplicationController"), clientsetadapter.FromUnversionedClient(restClient))
+func StopRC(rc *api.ReplicationController, clientset clientset.Interface) error {
+	reaper, err := kubectl.ReaperFor(api.Kind("ReplicationController"), clientset)
 	if err != nil || reaper == nil {
 		return err
 	}
@@ -313,8 +311,8 @@ func StopRC(rc *api.ReplicationController, restClient *client.Client) error {
 }
 
 // ScaleRC scales the given rc to the given replicas.
-func ScaleRC(name, ns string, replicas int32, restClient *client.Client) (*api.ReplicationController, error) {
-	scaler, err := kubectl.ScalerFor(api.Kind("ReplicationController"), clientsetadapter.FromUnversionedClient(restClient))
+func ScaleRC(name, ns string, replicas int32, clientset clientset.Interface) (*api.ReplicationController, error) {
+	scaler, err := kubectl.ScalerFor(api.Kind("ReplicationController"), clientset)
 	if err != nil {
 		return nil, err
 	}
@@ -324,64 +322,13 @@ func ScaleRC(name, ns string, replicas int32, restClient *client.Client) (*api.R
 	if err != nil {
 		return nil, err
 	}
-	scaled, err := restClient.ReplicationControllers(ns).Get(name)
+	scaled, err := clientset.Core().ReplicationControllers(ns).Get(name)
 	if err != nil {
 		return nil, err
 	}
 	return scaled, nil
 }
 
-// StartRC creates given rc if it doesn't already exist, then updates it via kubectl's scaler.
-func StartRC(controller *api.ReplicationController, restClient *client.Client) (*api.ReplicationController, error) {
-	created, err := restClient.ReplicationControllers(controller.Namespace).Get(controller.Name)
-	if err != nil {
-		glog.Infof("Rc %v doesn't exist, creating", controller.Name)
-		created, err = restClient.ReplicationControllers(controller.Namespace).Create(controller)
-		if err != nil {
-			return nil, err
-		}
-	}
-	// If we just created an rc, wait till it creates its replicas.
-	return ScaleRC(created.Name, created.Namespace, controller.Spec.Replicas, restClient)
-}
-
-// StartPods check for numPods in namespace. If they exist, it no-ops, otherwise it starts up
-// a temp rc, scales it to match numPods, then deletes the rc leaving behind the pods.
-func StartPods(namespace string, numPods int, host string, restClient *client.Client) error {
-	start := time.Now()
-	defer func() {
-		glog.Infof("StartPods took %v with numPods %d", time.Since(start), numPods)
-	}()
-	hostField := fields.OneTermEqualSelector(api.PodHostField, host)
-	options := api.ListOptions{FieldSelector: hostField}
-	pods, err := restClient.Pods(namespace).List(options)
-	if err != nil || len(pods.Items) == numPods {
-		return err
-	}
-	glog.Infof("Found %d pods that match host %v, require %d", len(pods.Items), hostField, numPods)
-	// For the sake of simplicity, assume all pods in namespace have selectors matching TestRCManifest.
-	controller := RCFromManifest(TestRCManifest)
-
-	// Overwrite namespace
-	controller.ObjectMeta.Namespace = namespace
-	controller.Spec.Template.ObjectMeta.Namespace = namespace
-
-	// Make the rc unique to the given host.
-	controller.Spec.Replicas = int32(numPods)
-	controller.Spec.Template.Spec.NodeName = host
-	controller.Name = controller.Name + host
-	controller.Spec.Selector["host"] = host
-	controller.Spec.Template.Labels["host"] = host
-
-	if rc, err := StartRC(controller, restClient); err != nil {
-		return err
-	} else {
-		// Delete the rc, otherwise when we restart master components for the next benchmark
-		// the rc controller will race with the pods controller in the rc manager.
-		return restClient.ReplicationControllers(namespace).Delete(rc.Name, nil)
-	}
-}
-
 func RunAMaster(masterConfig *master.Config) (*master.Master, *httptest.Server) {
 	if masterConfig == nil {
 		masterConfig = NewMasterConfig()