update

src/evaluator.go

@@ -14,6 +14,8 @@ type Evaluator struct {
 	factorNode   float64
 	factorRack   float64
 	factorDomain float64
+
+	costLoad float64
 }
 
 func (eva *Evaluator) init(nodes []NodeStatus, tasks []Task) {
@@ -28,6 +30,7 @@ func (eva *Evaluator) init(nodes []NodeStatus, tasks []Task) {
 	eva.factorDomain = 40.0
 	eva.cost = 0.0
 	eva.costNetwork = 0.0
+	eva.costLoad = 0.0
 }
 
 func (eva *Evaluator) add(node NodeStatus, task Task) {
@@ -63,6 +66,20 @@ func (eva *Evaluator) add(node NodeStatus, task Task) {
 		eva.totalWorker++
 	}
 	eva.cost = eva.costNetwork
+
+	if task.IsPS {
+		//eva.costLoad += 1
+	} else {
+		//eva.costLoad += 0.5
+	}
+	numberGPU := 1
+	for _, gpu := range node.Status {
+		if gpu.MemoryAllocated != 0 {
+			numberGPU += 1
+		}
+	}
+	eva.costLoad += float64(numberGPU) / float64(len(node.Status))
+
 }
 
 func (eva *Evaluator) remove(node NodeStatus, task Task) {
@@ -88,10 +105,23 @@ func (eva *Evaluator) remove(node NodeStatus, task Task) {
 		eva.totalWorker--
 	}
 	eva.cost = eva.costNetwork
+
+	if task.IsPS {
+		//eva.costLoad -= 1
+	} else {
+		//eva.costLoad -= 0.5
+	}
+	numberGPU := 1
+	for _, gpu := range node.Status {
+		if gpu.MemoryAllocated != 0 {
+			numberGPU += 1
+		}
+	}
+	eva.costLoad -= float64(numberGPU) / float64(len(node.Status))
 }
 
 func (eva *Evaluator) calculate() float64 {
-	return eva.cost
+	return eva.cost + eva.costLoad/float64(eva.totalPS+eva.totalWorker)
 }
 
 func evaluate(allocation Allocation) float64 {
@@ -189,6 +219,7 @@ func evaluate(allocation Allocation) float64 {
 	costLB *= 100
 	//fmt.Println(costLB)
 
-	cost := 0.0*costLB + 1.0*costNetwork
+	cost := costNetwork
+	//cost := 0.0*costLB + 1.0*costNetwork
 	return cost
 }

src/ga.go

@@ -3,14 +3,10 @@ package main
 import (
 	"math/rand"
 	"github.com/MaxHalford/eaopt"
-	"time"
-		"math"
+	"math"
 	log "github.com/sirupsen/logrus"
 )
 
-var nodesMap map[string]NodeStatus
-var tasksMap map[string]Task
-
 // A resource allocation
 type Allocation struct {
 	TasksOnNode map[string][]Task // tasks on nodes[id]
@@ -18,6 +14,7 @@ type Allocation struct {
 	NodeIDs     []string
 	Flags       map[string]bool
 	Evaluator   Evaluator
+	Tasks       []Task
 }
 
 func randomFit(allocation Allocation, task Task) (string, bool) {
@@ -27,10 +24,10 @@ func randomFit(allocation Allocation, task Task) (string, bool) {
 		numberGPU := 0
 		for _, gpu := range allocation.Nodes[nodeID].Status {
 			if gpu.MemoryAllocated == 0 {
-				numberGPU += 0
+				numberGPU += 1
 			}
 		}
-		if _, ok := allocation.Nodes[nodeID]; ok && len(allocation.TasksOnNode[nodeID]) < numberGPU {
+		if task.NumberGPU <= numberGPU {
 			flag = true
 			break
 		}
@@ -42,13 +39,16 @@ func firstFit(allocation Allocation, task Task) (string, bool) {
 	flag := false
 	nodeID := ""
 	for _, nodeID = range allocation.NodeIDs {
+		if _, ok := allocation.Nodes[nodeID]; !ok {
+			continue
+		}
 		numberGPU := 0
 		for _, gpu := range allocation.Nodes[nodeID].Status {
 			if gpu.MemoryAllocated == 0 {
-				numberGPU += 0
+				numberGPU += 1
 			}
 		}
-		if _, ok := allocation.Nodes[nodeID]; ok && len(allocation.TasksOnNode[nodeID]) < numberGPU {
+		if task.NumberGPU <= numberGPU {
 			flag = true
 			break
 		}
@@ -57,6 +57,8 @@ func firstFit(allocation Allocation, task Task) (string, bool) {
 }
 
 func fastBestFit(nodes []NodeStatus, tasks []Task) Allocation {
+	//log.Info(nodes)
+	//log.Info(tasks)
 	eva := Evaluator{}
 	eva.init(nodes, tasks)
 
@@ -64,90 +66,76 @@ func fastBestFit(nodes []NodeStatus, tasks []Task) Allocation {
 	allocation.TasksOnNode = map[string][]Task{}
 	for _, task := range tasks {
 		minCost := math.MaxFloat64
-		nodeID := ""
-		for _, node := range nodes {
+		var best *NodeStatus
+		for i, node := range nodes {
 			if _, ok := allocation.TasksOnNode[node.ClientID]; !ok {
 				allocation.TasksOnNode[node.ClientID] = []Task{}
 			}
 			numberGPU := 0
-			for _, gpu := range allocation.Nodes[nodeID].Status {
+			for _, gpu := range node.Status {
 				if gpu.MemoryAllocated == 0 {
-					numberGPU += 0
+					numberGPU += 1
 				}
 			}
-			if len(allocation.TasksOnNode[node.ClientID]) >= numberGPU {
+			if task.NumberGPU > numberGPU {
 				continue
 			}
 			eva.add(node, task)
 			cost := eva.calculate()
 			eva.remove(node, task)
-			if cost < minCost || nodeID == "" {
+			//log.Info(node, cost)
+			if cost < minCost || best == nil {
 				minCost = cost
-				nodeID = node.ClientID
+				best = &nodes[i]
 			}
-			//fmt.Println(cost)
 		}
-		if nodeID == "" {
+		log.Info(task, " choose ", best.ClientID)
+		if best == nil {
 			allocation.Flags["valid"] = false
 			break
 		} else {
 			//fmt.Println(task, nodeID, allocation.TasksOnNode, minCost)
-			allocation.TasksOnNode[nodeID] = append(allocation.TasksOnNode[nodeID], task)
-			eva.add(nodesMap[nodeID], task)
-		}
-	}
-	log.Println(eva.calculate())
-	return allocation
-}
-
-func bestFit(allocation Allocation, task Task) (string, bool) {
-	flag := false
-	nodeID := ""
-	minCost := math.MaxFloat64
-	for _, id := range allocation.NodeIDs {
-		numberGPU := 0
-		for _, gpu := range allocation.Nodes[id].Status {
-			if gpu.MemoryAllocated == 0 {
-				numberGPU += 0
-			}
-		}
-		if _, ok := allocation.Nodes[id]; ok && len(allocation.TasksOnNode[id]) < numberGPU {
-			/* add */
-			allocation.TasksOnNode[id] = append(allocation.TasksOnNode[id], task)
-
-			/* evaluate */
-			cost := evaluate(allocation)
-
-			/* revert */
-			idx := -1
-			for i, task2 := range allocation.TasksOnNode[id] {
-				if task2.Name == task.Name {
-					idx = i
+			allocation.TasksOnNode[best.ClientID] = append(allocation.TasksOnNode[best.ClientID], task)
+			eva.add(*best, task)
+			for i := range best.Status {
+				//allocate more than 1
+				if best.Status[i].MemoryAllocated == 0 {
+					best.Status[i].MemoryAllocated += task.MemoryGPU
+					break
 				}
 			}
-			copy(allocation.TasksOnNode[id][idx:], allocation.TasksOnNode[id][idx+1:])
-			allocation.TasksOnNode[id] = allocation.TasksOnNode[id][:len(allocation.TasksOnNode[id])-1]
-
-			if cost < minCost || !flag {
-				nodeID = id
-				minCost = cost
-			}
-			flag = true
 		}
 	}
-	return nodeID, flag
+	//log.Info(allocation.TasksOnNode)
+	log.Println("BestFit Cost:", eva.calculate())
+	return allocation
 }
 
 /* Evaluate the allocation */
 func (X Allocation) Evaluate() (float64, error) {
+	//log.Info(X)
 	if !X.Flags["valid"] {
 		//fmt.Println("Invalid allocation")
 		return math.MaxFloat64, nil
 	}
 
-	costNetwork := evaluate(X)
+	//costNetwork := evaluate(X)
 
-	cost := costNetwork
+	var nodes []NodeStatus
+	for _, node := range X.Nodes {
+		nodes = append(nodes, node)
+	}
+
+	eva := Evaluator{}
+	eva.init(nodes, X.Tasks)
+	for node, tasks := range X.TasksOnNode {
+		for _, task := range tasks {
+			eva.add(X.Nodes[node], task)
+		}
+	}
+
+	cost := eva.calculate()
+	//log.Info(cost)
 	//fmt.Println(taskToNode, cost, len(X.Nodes))
 	return float64(cost), nil
 }
@@ -189,6 +177,12 @@ func (X Allocation) Mutate(rng *rand.Rand) {
 		for _, task := range tasks {
 			if nodeID, ok := firstFit(X, task); ok {
 				X.TasksOnNode[nodeID] = append(X.TasksOnNode[nodeID], task)
+				for i := range X.Nodes[nodeID].Status {
+					if X.Nodes[nodeID].Status[i].MemoryAllocated == 0 {
+						X.Nodes[nodeID].Status[i].MemoryAllocated += task.MemoryGPU
+						break
+					}
+				}
 			} else {
 				X.Flags["valid"] = false
 			}
@@ -196,6 +190,7 @@ func (X Allocation) Mutate(rng *rand.Rand) {
 	}
 	//fmt.Println("After", X)
 
+	return
 	/* move tasks */
 	if !X.Flags["valid"] {
 		//fmt.Println("Invalid allocation")
@@ -230,7 +225,6 @@ func (X Allocation) Crossover(Y eaopt.Genome, rng *rand.Rand) {
 		if !Y.(Allocation).Flags["valid"] || !X.Flags["valid"] {
 			return
 		}
-		//fmt.Println("Crossover")
 		taskToNode := map[string]string{}
 		for nodeID, tasks := range X.TasksOnNode {
 			for _, task := range tasks {
@@ -265,6 +259,12 @@ func (X Allocation) Crossover(Y eaopt.Genome, rng *rand.Rand) {
 			//fmt.Println(X.TasksOnNode)
 			copy(X.TasksOnNode[nodeID2][idx:], X.TasksOnNode[nodeID2][idx+1:])
 			X.TasksOnNode[nodeID2] = X.TasksOnNode[nodeID2][:len(X.TasksOnNode[nodeID2])-1]
+			for i := range X.Nodes[nodeID].Status {
+				if X.Nodes[nodeID].Status[i].MemoryAllocated == 0 {
+					X.Nodes[nodeID].Status[i].MemoryAllocated -= task.MemoryGPU
+					break
+				}
+			}
 			//fmt.Println(X.TasksOnNode)
 		}
 		/* reschedule tasks on tgt node */
@@ -291,6 +291,12 @@ func (X Allocation) Crossover(Y eaopt.Genome, rng *rand.Rand) {
 		for _, task := range tasks {
 			if nodeID, ok := firstFit(X, task); ok {
 				X.TasksOnNode[nodeID] = append(X.TasksOnNode[nodeID], task)
+				for i := range X.Nodes[nodeID].Status {
+					if X.Nodes[nodeID].Status[i].MemoryAllocated == 0 {
+						X.Nodes[nodeID].Status[i].MemoryAllocated += task.MemoryGPU
+						break
+					}
+				}
 			} else {
 				X.Flags["valid"] = false
 			}
@@ -319,74 +325,3 @@ func (X Allocation) Clone() eaopt.Genome {
 	}
 	return Y
 }
-
-func VectorFactory(rng *rand.Rand) eaopt.Genome {
-	allocation := Allocation{TasksOnNode: map[string][]Task{}, Nodes: map[string]NodeStatus{}, Flags: map[string]bool{"valid": true}}
-
-	var nodes []NodeStatus
-	var tasks []Task
-
-	for _, node := range nodesMap {
-		nodes = append(nodes, node)
-	}
-	for _, task := range tasksMap {
-		tasks = append(tasks, task)
-	}
-
-	/* shuffle */
-	for n := len(nodes); n > 0; n-- {
-		randIndex := rng.Intn(n)
-		nodes[n-1], nodes[randIndex] = nodes[randIndex], nodes[n-1]
-	}
-	for n := len(tasks); n > 0; n-- {
-		randIndex := rng.Intn(n)
-		tasks[n-1], tasks[randIndex] = tasks[randIndex], tasks[n-1]
-	}
-
-	/* pick nodes */
-	for _, node := range nodesMap {
-		allocation.Nodes[node.ClientID] = node
-		allocation.NodeIDs = append(allocation.NodeIDs, node.ClientID)
-	}
-
-	t := rng.Int() % 10
-	if t == -1 {
-		/* best-fit */
-		ts := time.Now()
-
-		/*
-		for _, task := range tasks {
-			if nodeID, ok := bestFit(allocation, task); ok {
-				allocation.TasksOnNode[nodeID] = append(allocation.TasksOnNode[nodeID], task)
-			} else {
-				allocation.Flags["valid"] = false
-			}
-		}
-		*/
-
-		allocation.TasksOnNode = fastBestFit(nodes, tasks).TasksOnNode
-		log.Println(time.Since(ts))
-		//fmt.Println("Best Fit")
-	} else if t%2 == 0 {
-		/* first-fit */
-		for _, task := range tasks {
-			if nodeID, ok := randomFit(allocation, task); ok {
-				allocation.TasksOnNode[nodeID] = append(allocation.TasksOnNode[nodeID], task)
-			} else {
-				allocation.Flags["valid"] = false
-			}
-		}
-	} else {
-		/* random-fit */
-		for _, task := range tasks {
-			if nodeID, ok := randomFit(allocation, task); ok {
-				allocation.TasksOnNode[nodeID] = append(allocation.TasksOnNode[nodeID], task)
-			} else {
-				allocation.Flags["valid"] = false
-			}
-		}
-	}
-	//fmt.Println(evaluatue(allocation))
-	//fmt.Println(allocation)
-	return allocation
-}

src/ga_test.go

@@ -4,53 +4,61 @@ import (
 	"strconv"
 	"math/rand"
 	"time"
-	"log"
+	log "github.com/sirupsen/logrus"
 	"github.com/MaxHalford/eaopt"
 	"math"
 	"testing"
 )
 
-func TestGA(t *testing.T) {
-	numTask := 20
-
-	nodesMap = map[string]NodeStatus{}
-	tasksMap = map[string]Task{}
-
-	for i := 0; i < numTask*3; i++ {
-		node := NodeStatus{ClientID: strconv.Itoa(i), Rack: strconv.Itoa(i % 40), Domain: strconv.Itoa(i % 4)}
-		node.NumCPU = 24
-		node.MemTotal = 188
-		node.TotalBW = 100
-		cnt := rand.Intn(3) + 1
-		for i := 0; i < cnt; i++ {
-			node.Status = append(node.Status, GPUStatus{MemoryTotal: 11439, MemoryAllocated: 0, UUID: node.ClientID + strconv.Itoa(i)})
-		}
-		nodesMap[strconv.Itoa(i)] = node
-	}
-	for i := 0; i < numTask; i++ {
-		isPS := false
-		if i >= 3 {
-			isPS = true
-		}
-		task := Task{Name: strconv.Itoa(i), IsPS: isPS}
-		task.Memory = 4
-		task.NumberCPU = 2
-		task.NumberGPU = 1
-		tasksMap[strconv.Itoa(i)] = task
-	}
+func TgenerateCase() ([]NodeStatus, []Task) {
+	numTask := 6
 
 	var nodes []NodeStatus
 	var tasks []Task
 
-	for _, node := range nodesMap {
+	for i := 0; i < numTask*3; i++ {
+		node := NodeStatus{ClientID: strconv.Itoa(i), Rack: "Rack-" + strconv.Itoa(i%40), Domain: "Domain-" + strconv.Itoa(i%4)}
+		node.NumCPU = 24
+		node.UtilCPU = 2.0
+		node.MemTotal = 188
+		node.MemAvailable = 20
+		node.TotalBW = 100
+		//cnt := 4
+		cnt := rand.Intn(3) + 1
+		for i := 0; i < cnt; i++ {
+			node.Status = append(node.Status, GPUStatus{MemoryTotal: 11439, MemoryAllocated: 0, UUID: node.ClientID + "-" + strconv.Itoa(i)})
+		}
 		nodes = append(nodes, node)
 	}
-	for _, task := range tasksMap {
+	for i := 0; i < numTask; i++ {
+		isPS := false
+		if i%4 == 0 {
+			isPS = true
+		}
+		task := Task{Name: "task-" + strconv.Itoa(i), IsPS: isPS}
+		task.Memory = 4
+		task.NumberCPU = 2
+		task.NumberGPU = 1
+		task.MemoryGPU = 4096
 		tasks = append(tasks, task)
 	}
+	return nodes, tasks
+}
+
+func TestBestFit(t *testing.T) {
+	nodes, tasks := TgenerateCase()
+	for _, node := range nodes {
+		log.Info(node)
+	}
 	s := time.Now()
 	allocation := fastBestFit(nodes, tasks)
 	log.Println(time.Since(s))
+	log.Println(allocation)
+}
+
+func TestGA(t *testing.T) {
+	return
+	nodes, tasks := TgenerateCase()
 
 	// Instantiate a GA with a GAConfig
 	var ga, err = eaopt.NewDefaultGAConfig().NewGA()
@@ -62,7 +70,7 @@ func TestGA(t *testing.T) {
 	// Set the number of generations to run for
 	ga.NGenerations = math.MaxInt32
 	ga.NPops = 1
-	ga.PopSize = 30 + uint(numTask/2)
+	ga.PopSize = 30 + uint(len(tasks)/2)
 
 	// Add a custom print function to track progress
 	ga.Callback = func(ga *eaopt.GA) {
@@ -90,15 +98,92 @@ func TestGA(t *testing.T) {
 		return false
 	}
 
+	var f = func(rng *rand.Rand) eaopt.Genome {
+		allocation := Allocation{TasksOnNode: map[string][]Task{}, Nodes: map[string]NodeStatus{}, Flags: map[string]bool{"valid": true}}
+
+		//log.Println(nodes)
+		var nodesT []NodeStatus
+		for _, node := range nodes {
+			nodesT = append(nodesT, node.Copy())
+		}
+
+		//nodesT[0].Status[0].MemoryAllocated = 100
+		//log.Println(nodes[0].Status[0].MemoryAllocated)
+
+		//log.Println(&nodesT[0])
+		//log.Println(&nodes[0])
+
+		for _, node := range nodesT {
+			allocation.Nodes[node.ClientID] = node
+		}
+		for _, task := range tasks {
+			allocation.Tasks = append(allocation.Tasks, task)
+		}
+
+		/* shuffle */
+		for n := len(tasks); n > 0; n-- {
+			randIndex := rng.Intn(n)
+			allocation.Tasks[n-1], allocation.Tasks[randIndex] = allocation.Tasks[randIndex], allocation.Tasks[n-1]
+		}
+
+		/* pick nodes */
+		for _, node := range nodesT {
+			allocation.Nodes[node.ClientID] = node
+			allocation.NodeIDs = append(allocation.NodeIDs, node.ClientID)
+		}
+
+		t := rng.Int() % 10
+		if t == 0 {
+			/* best-fit */
+			ts := time.Now()
+			allocation.TasksOnNode = fastBestFit(nodesT, tasks).TasksOnNode
+			log.Println(time.Since(ts))
+			//fmt.Println("Best Fit")
+		} else if t%2 == 0 {
+			/* first-fit */
+			for _, task := range tasks {
+				if nodeID, ok := randomFit(allocation, task); ok {
+					allocation.TasksOnNode[nodeID] = append(allocation.TasksOnNode[nodeID], task)
+					for i := range allocation.Nodes[nodeID].Status {
+						if allocation.Nodes[nodeID].Status[i].MemoryAllocated == 0 {
+							allocation.Nodes[nodeID].Status[i].MemoryAllocated += task.MemoryGPU
+							break
+						}
+					}
+				} else {
+					allocation.Flags["valid"] = false
+				}
+			}
+		} else {
+			/* random-fit */
+			for _, task := range tasks {
+				if nodeID, ok := randomFit(allocation, task); ok {
+					allocation.TasksOnNode[nodeID] = append(allocation.TasksOnNode[nodeID], task)
+					for i := range allocation.Nodes[nodeID].Status {
+						if allocation.Nodes[nodeID].Status[i].MemoryAllocated == 0 {
+							allocation.Nodes[nodeID].Status[i].MemoryAllocated += task.MemoryGPU
+							break
+						}
+					}
+				} else {
+					allocation.Flags["valid"] = false
+				}
+			}
+		}
+		//fmt.Println(evaluatue(allocation))
+		//fmt.Println(allocation)
+		return allocation
+
+	}
+
 	// Find the minimum
-	err = ga.Minimize(VectorFactory)
+	err = ga.Minimize(f)
 	log.Println(time.Since(ts))
 	log.Println(ga.HallOfFame[0].Genome.(Allocation).TasksOnNode)
-	//fmt.Println(ga.HallOfFame[0].Genome.(Allocation).Nodes)
+	//log.Println(ga.HallOfFame[0].Genome.(Allocation).Flags)
+	//log.Println(ga.HallOfFame[0].Genome.(Allocation).Nodes)
 	if err != nil {
 		log.Println(err)
 		return
 	}
-
-	log.Println(allocation)
 }

src/pool_status.go

@@ -40,3 +40,10 @@ type NodeStatus struct {
 	TotalBW      float64     `json:"bw_total"`
 	Status       []GPUStatus `json:"status"`
 }
+
+func (X *NodeStatus) Copy() NodeStatus {
+	res := *X
+	res.Status = make([]GPUStatus, len(X.Status))
+	copy(res.Status, X.Status)
+	return res
+}

src/resource_pool.go

@@ -158,14 +158,14 @@ func (pool *ResourcePool) checkDeadNodes() {
 
 func (pool *ResourcePool) GPUModelToPower(model string) int {
 	mapper := map[string]int{
-		"K40": 1, "Tesla K40": 1,
-		"K80": 2, "Tesla K80": 2,
-		"P100": 3, "Tesla P100": 3,
+		"K40": 2, "Tesla K40": 2,
+		"K80": 3, "Tesla K80": 3,
+		"P100": 4, "Tesla P100": 4,
 	}
 	if power, err := mapper[model]; !err {
 		return power
 	}
-	return 0
+	return 1
 }
 
 func (pool *ResourcePool) getNodePool(name string) int {
@@ -639,12 +639,16 @@ func (pool *ResourcePool) acquireResource(job Job) []NodeStatus {
 	locks := map[int]*sync.Mutex{}
 
 	allocationType := 0
-	availableGPUs := map[string][]GPUStatus{}
 
-	var candidates []*NodeStatus
+	var candidates []NodeStatus
+
+	if pool.TotalGPU == 0 {
+		return []NodeStatus{}
+	}
+	loadRatio := float64(pool.UsingGPU) / float64(pool.TotalGPU)
 
 	/* first, choose sharable GPUs */
-	if pool.enableShare && (pool.TotalGPU != 0 && len(job.Tasks) == 1 && task.NumberGPU == 1 && float64(pool.UsingGPU)/float64(pool.TotalGPU) >= pool.enableShareRatio) {
+	if pool.enableShare && len(job.Tasks) == 1 && task.NumberGPU == 1 && loadRatio >= pool.enableShareRatio {
 		// check sharable
 		allocationType = 1
 		if util, valid := InstanceOfOptimizer().predictUtilGPU(job.Name); valid {
@@ -671,13 +675,12 @@ func (pool *ResourcePool) acquireResource(job Job) []NodeStatus {
 								}
 								if totalUtil < 100 {
 									available = append(available, status)
-									availableGPUs[node.ClientID] = available
 								}
 							}
 						}
 					}
 					if len(available) >= task.NumberGPU {
-						candidates = append(candidates, node)
+						candidates = append(candidates, *node)
 						if len(candidates) >= len(job.Tasks)*3+5 {
 							break
 						}
@@ -711,8 +714,7 @@ func (pool *ResourcePool) acquireResource(job Job) []NodeStatus {
 					}
 				}
 				if len(available) >= task.NumberGPU {
-					candidates = append(candidates, node)
-					availableGPUs[node.ClientID] = available
+					candidates = append(candidates, *node)
 					if len(candidates) >= len(job.Tasks)*3+5 {
 						break
 					}
@@ -733,11 +735,7 @@ func (pool *ResourcePool) acquireResource(job Job) []NodeStatus {
 	if len(candidates) == 0 && len(job.Tasks) == 1 && task.NumberGPU == 1 && pool.enablePreSchedule {
 		estimate, valid := InstanceOfOptimizer().predictTime(job.Name)
 
-		//log.Info(pool.TotalGPU)
-		//log.Info(estimate, valid)
-		//log.Info(scheduler.UsingGPU)
-
-		if pool.TotalGPU != 0 && float64(pool.UsingGPU)/float64(pool.TotalGPU) >= pool.enablePreScheduleRatio && valid {
+		if loadRatio >= pool.enablePreScheduleRatio && valid {
 			allocationType = 3
 			for cur := start; ; {
 				if _, ok := locks[cur.ID]; !ok {
@@ -765,8 +763,7 @@ func (pool *ResourcePool) acquireResource(job Job) []NodeStatus {
 						}
 					}
 					if len(available) >= task.NumberGPU {
-						candidates = append(candidates, node)
-						availableGPUs[node.ClientID] = available
+						candidates = append(candidates, *node)
 						if len(candidates) >= len(job.Tasks)*3+5 {
 							break
 						}
@@ -792,44 +789,69 @@ func (pool *ResourcePool) acquireResource(job Job) []NodeStatus {
 	/* assign */
 	var ress []NodeStatus
 	if len(candidates) > 0 {
-		/*
-				for range job.Tasks { //append would cause uncertain order
-					resources = append(resources, NodeStatus{ClientID: "null"})
-				}
-				*/
-
-		var nodes []NodeStatus
-		if len(job.Tasks) == 1 {
-			node := pool.pickNode(candidates, availableGPUs, task, job, []NodeStatus{})
-			nodes = append(nodes, *node)
+		for range job.Tasks { //append would cause uncertain order
+			ress = append(ress, NodeStatus{ClientID: "null"})
 		}
 
-		for _, node := range nodes {
-			res := NodeStatus{}
-			res.ClientID = node.ClientID
-			res.ClientHost = node.ClientHost
-			res.Status = availableGPUs[node.ClientID][0:task.NumberGPU]
-			res.NumCPU = task.NumberCPU
-			res.MemTotal = task.Memory
+		var nodesT []NodeStatus
+		for _, node := range candidates {
+			nodesT = append(nodesT, node.Copy())
+		}
 
-			for i := range res.Status {
-				for j := range node.Status {
-					if res.Status[i].UUID == node.Status[j].UUID {
-						if node.Status[j].MemoryAllocated == 0 {
-							pool.UsingGPUMu.Lock()
-							pool.UsingGPU ++
-							pool.UsingGPUMu.Unlock()
+		allocation := fastBestFit(nodesT, job.Tasks)
+		if !allocation.Flags["valid"] {
+			return []NodeStatus{}
+		}
+
+		for nodeID, tasks := range allocation.TasksOnNode {
+			var node *NodeStatus
+			for i := range candidates {
+				if candidates[i].ClientID == nodeID {
+					node = &candidates[i]
+				}
+			}
+
+			var available []GPUStatus
+			for _, gpu := range node.Status {
+				if gpu.MemoryAllocated == 0 {
+					available = append(available, gpu)
+				}
+			}
+			for _, task := range tasks {
+				res := NodeStatus{}
+				res.ClientID = node.ClientID
+				res.ClientHost = node.ClientHost
+				res.NumCPU = task.NumberCPU
+				res.MemTotal = task.Memory
+				res.Status = available[0:task.NumberGPU]
+				available = available[task.NumberGPU:]
+
+				for i := range res.Status {
+					for j := range node.Status {
+						if res.Status[i].UUID == node.Status[j].UUID {
+							if node.Status[j].MemoryAllocated == 0 {
+								pool.UsingGPUMu.Lock()
+								pool.UsingGPU ++
+								pool.UsingGPUMu.Unlock()
+							}
+							node.Status[j].MemoryAllocated += task.MemoryGPU
+							res.Status[i].MemoryTotal = task.MemoryGPU
 						}
-						node.Status[j].MemoryAllocated += task.MemoryGPU
-						res.Status[i].MemoryTotal = task.MemoryGPU
 					}
 				}
+				for _, t := range res.Status {
+					pool.attach(t.UUID, job.Name)
+				}
+
+				for i := range job.Tasks {
+					if job.Tasks[i].Name == task.Name {
+						ress[i] = res
+					}
+				}
+
 			}
-			for _, t := range res.Status {
-				pool.attach(t.UUID, job.Name)
-			}
-			ress = append(ress, res)
 		}
+
 	}
 
 	for segID, lock := range locks {