diff --git a/src/job_manager.go b/src/job_manager.go
index b1b297b..c810347 100644
--- a/src/job_manager.go
+++ b/src/job_manager.go
@@ -343,6 +343,7 @@ func (jm *JobManager) stop(force bool) MsgStop {
 			if res.Code != 0 {
 				log.Warn(res.Error)
 			}
+			log.Info(task.Id, " is killed")
 
 		}(taskStatus)
 	}
diff --git a/src/resource_pool.go b/src/resource_pool.go
index 057ab5b..1a50b38 100644
--- a/src/resource_pool.go
+++ b/src/resource_pool.go
@@ -920,6 +920,7 @@ func (pool *ResourcePool) releaseResource(job Job, agent NodeStatus) {
 				node.Status[j].MemoryAllocated -= gpu.MemoryTotal
 				if node.Status[j].MemoryAllocated < 0 {
 					// in case of error
+					/* Case 0: a node is offline and then online, the allocation info will be lost */
 					log.Warn(node.ClientID, " UUID=", gpu.UUID, " More Memory Allocated")
 					node.Status[j].MemoryAllocated = 0
 				}
diff --git a/src/scheduler_fair.go b/src/scheduler_fair.go
index 938e7c8..d9d23ed 100644
--- a/src/scheduler_fair.go
+++ b/src/scheduler_fair.go
@@ -16,6 +16,8 @@ type SchedulerFair struct {
 	queues   map[string][]Job
 	queuesMu sync.Mutex
 
+	enableBorrow  bool
+	IOUs          map[string]map[string]*ResourceCount
 	queuesQuota   map[string]*ResourceCount
 	queuesQuotaMu sync.Mutex
 
@@ -39,6 +41,8 @@ func (scheduler *SchedulerFair) Start() {
 	scheduler.history = []*Job{}
 	scheduler.queues = map[string][]Job{}
 	scheduler.queues["default"] = []Job{}
+	scheduler.enableBorrow = true
+	scheduler.IOUs = map[string]map[string]*ResourceCount{}
 	scheduler.queuesQuota = map[string]*ResourceCount{}
 	scheduler.resourceAllocations = map[string]*ResourceCount{}
 	scheduler.enabled = true
@@ -69,38 +73,102 @@ func (scheduler *SchedulerFair) Start() {
 
 			scheduler.queuesMu.Lock()
 			scheduler.queuesQuotaMu.Lock()
-
-			go func() {
-				scheduler.UpdateQuota()
-			}()
-
+			/* choose queue which has the largest job */
 			bestQueue := ""
-			numberGPU := math.MaxInt64
-			//numberCPU := math.MaxInt64
-			/* phase 1 */
+			maxNumberGPU := math.MaxInt64
+			maxNumberCPU := math.MaxInt64
+			/* phase 1: execute jobs using self quota */
 			for queue, jobs := range scheduler.queues {
-				/* find smallest job */
+				/* find largest job */
 				if len(jobs) > 0 {
+					/* calculate resource request of head job */
 					numberGPUtmp := 0
 					numberCPUtmp := 0
 					for _, task := range jobs[0].Tasks {
 						numberGPUtmp += task.NumberGPU
 						numberCPUtmp += task.NumberCPU
 					}
+					/* if queue quota cannot satisfy, skip */
 					if quota, ok := scheduler.queuesQuota[queue]; !ok || quota.NumberGPU/1000 < numberGPUtmp {
 						continue
 					}
-					if bestQueue == "" || numberGPUtmp < numberGPU || (numberGPUtmp == numberGPU) {
+					/* the more, the better */
+					if bestQueue == "" || numberGPUtmp > maxNumberGPU || (numberGPUtmp == maxNumberGPU && numberCPUtmp > maxNumberCPU) {
+						/* cannot borrow more if already borrowed */
+						if _, ok := scheduler.IOUs[queue]; ok && len(scheduler.IOUs[queue]) > 0 {
+							continue
+						}
 						bestQueue = queue
-						numberGPU = numberGPUtmp
-						//numberCPU = numberCPUtmp
+						maxNumberGPU = numberGPUtmp
+						maxNumberCPU = numberCPUtmp
 					}
 				}
 			}
 
-			/* phase 2 */
-			if bestQueue == "" {
+			/* phase 2: borrow */
+			if bestQueue == "" && scheduler.enableBorrow {
+				/* firstly, check if quota sum can run a job */
+				totalGPU := 0
+				for _, quota := range scheduler.queuesQuota {
+					totalGPU += quota.NumberGPU
+				}
+				/* find job which is short of least resource */
+				minRequestGPU := math.MaxInt32
+				for queue, jobs := range scheduler.queues {
+					if len(jobs) == 0 {
+						continue
+					}
+					numberGPUtmp := 0
+					for _, task := range jobs[0].Tasks {
+						numberGPUtmp += task.NumberGPU
+					}
+					if _, ok := scheduler.queuesQuota[queue]; !ok {
+						scheduler.queuesQuota[queue] = &ResourceCount{}
+					}
+					needGPU := numberGPUtmp*1000 - scheduler.queuesQuota[queue].NumberGPU
+					/* the less, the better */
+					if bestQueue == "" || needGPU < minRequestGPU {
+						bestQueue = queue
+						minRequestGPU = needGPU
+					}
+				}
+				/* if totalGPU can satisfy that job, start borrowing */
+				if bestQueue != "" && totalGPU >= minRequestGPU {
+					for {
+						/* if all satisfied, break */
+						if minRequestGPU == 0 {
+							break
+						}
+						least := math.MaxInt32
+						for queue, quota := range scheduler.queuesQuota {
+							if queue == bestQueue {
+								continue
+							}
+							if quota.NumberGPU < least {
+								least = quota.NumberGPU
+							}
+						}
+						if minRequestGPU < least*(len(scheduler.queuesQuota)-1) {
+							least = minRequestGPU / (len(scheduler.queuesQuota) - 1)
+						}
+						/* start borrow */
+						for queue, quota := range scheduler.queuesQuota {
+							quota.NumberGPU -= least
+							if _, ok := scheduler.IOUs[bestQueue]; !ok {
+								scheduler.IOUs[bestQueue] = map[string]*ResourceCount{}
+							}
+							IOU, ok := scheduler.IOUs[bestQueue][queue]
+							if !ok {
+								scheduler.IOUs[bestQueue][queue] = &ResourceCount{}
+								IOU = scheduler.IOUs[bestQueue][queue]
+							}
+							IOU.NumberGPU += least
+							minRequestGPU -= least
+							log.Info(bestQueue, " borrow ", least, " from ", queue)
+						}
+					}
 
+				}
 			}
 
 			/* launch that job */
@@ -164,8 +232,10 @@ func (scheduler *SchedulerFair) Start() {
 				}
 			} else {
 				log.Debug("No more jobs to scheduling ", time.Now())
+				go func() {
+					scheduler.UpdateQuota()
+				}()
 			}
-
 			scheduler.queuesQuotaMu.Unlock()
 			scheduler.queuesMu.Unlock()
 		}