diff --git a/src/job_manager.go b/src/job_manager.go
index f5c53f7..767e435 100644
--- a/src/job_manager.go
+++ b/src/job_manager.go
@@ -115,7 +115,7 @@ func (jm *JobManager) start() {
 			v.Set("cpu_limit", strconv.Itoa(jm.job.Tasks[i].NumberCPU))
 			v.Set("network", network)
 			v.Set("should_wait", "1")
-			v.Set("output_dir", "/output/")
+			v.Set("output_dir", "/tmp/")
 			v.Set("hdfs_dir", "http://hdfs-master:50070/user/yao/output/"+jm.job.Name)
 			v.Set("gpu_mem", strconv.Itoa(jm.job.Tasks[i].MemoryGPU))
 
@@ -171,6 +171,14 @@ func (jm *JobManager) start() {
 				log.Info(jm.job.Name, "-", i, " ", res.Status[i].Status)
 				/* save logs etc. */
 
+				if exitCode, ok := res.Status[i].State["ExitCode"].(int); ok {
+					if exitCode != 0 {
+						log.Warn(jm.job.Name+"-"+jm.job.Tasks[i].Name+" exited unexpected, exitCode=", exitCode)
+						jm.killedFlag = true
+						jm.scheduler.UpdateProgress(jm.job, Failed)
+					}
+				}
+
 				/* remove exited containers */
 				//v := url.Values{}
 				//v.Set("id", res.Status[i].Id)
diff --git a/src/resource_pool.go b/src/resource_pool.go
index b55176a..a4253f4 100644
--- a/src/resource_pool.go
+++ b/src/resource_pool.go
@@ -41,10 +41,12 @@ type ResourcePool struct {
 }
 
 func (pool *ResourcePool) start() {
+	log.Info("RM started ")
+
 	pool.networks = map[string]bool{}
 	pool.networksFree = map[string]bool{}
-	pool.versions = map[string]float64{}
 
+	pool.versions = map[string]float64{}
 	pool.bindings = map[string]map[string]int{}
 	pool.utils = map[string][]UtilGPUTimeSeries{}
 
@@ -221,7 +223,7 @@ func (pool *ResourcePool) update(node NodeStatus) {
 		defer pool.bindingsMu.Unlock()
 		for _, gpu := range node.Status {
 			if _, ok := pool.bindings[gpu.UUID]; ok {
-				if len(pool.bindings[gpu.UUID]) == 1 {
+				if _, ok2 := pool.utils[gpu.UUID]; ok2 {
 					pool.utils[gpu.UUID] = append(pool.utils[gpu.UUID],
 						UtilGPUTimeSeries{Time: (int)(time.Now().Unix()), Util: gpu.UtilizationGPU})
 				}
@@ -244,7 +246,7 @@ func (pool *ResourcePool) update(node NodeStatus) {
 
 	status, ok := seg.Nodes[node.ClientID]
 	if ok {
-		/* remain allocation info */
+		/* keep allocation info */
 		for i, GPU := range status.Status {
 			if GPU.UUID == node.Status[i].UUID {
 				node.Status[i].MemoryAllocated = GPU.MemoryAllocated
@@ -433,15 +435,21 @@ func (pool *ResourcePool) attach(GPU string, job string) {
 	if _, ok := pool.utils[GPU]; !ok {
 		pool.utils[GPU] = []UtilGPUTimeSeries{}
 	}
+
+	if len(pool.bindings[GPU]) > 1 {
+		delete(pool.utils, GPU)
+	}
 }
 
 func (pool *ResourcePool) detach(GPU string, jobName string) {
 	pool.bindingsMu.Lock()
 	defer pool.bindingsMu.Unlock()
 	if _, ok := pool.bindings[GPU]; ok {
-		if len(pool.bindings[GPU]) == 1 {
-			InstanceOfOptimizer().feed(jobName, pool.utils[GPU])
-			pool.utils[GPU] = []UtilGPUTimeSeries{}
+		if _, ok2 := pool.utils[GPU]; ok2 {
+			if len(pool.bindings[GPU]) == 1 {
+				InstanceOfOptimizer().feed(jobName, pool.utils[GPU])
+			}
+			delete(pool.utils, GPU)
 		}
 	}
 
diff --git a/src/scheduler_fair.go b/src/scheduler_fair.go
index e5a3d6e..162a12e 100644
--- a/src/scheduler_fair.go
+++ b/src/scheduler_fair.go
@@ -85,6 +85,7 @@ func (scheduler *SchedulerFair) Start() {
 	scheduler.parallelism = 1
 
 	go func() {
+		/* fair scheduler */
 		flag := true
 		for {
 			log.Debug("Scheduling")
@@ -286,6 +287,14 @@ func (scheduler *SchedulerFair) UpdateProgress(job Job, state State) {
 			}
 		}
 		break
+	case Failed:
+		for i := range scheduler.history {
+			if scheduler.history[i].Name == job.Name {
+				scheduler.history[i].Status = Failed
+				scheduler.history[i].UpdatedAt = int(time.Now().Unix())
+			}
+		}
+		break
 	}
 }
 
diff --git a/src/state.go b/src/state.go
index c3f79f0..066ad28 100644
--- a/src/state.go
+++ b/src/state.go
@@ -13,4 +13,6 @@ const (
 	Stopped
 	// finished successfully
 	Finished
+
+	Failed
 )