/
analysis.go
235 lines (206 loc) · 9.34 KB
/
analysis.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
package simulator
import (
"math"
"sync"
log "github.com/sirupsen/logrus"
corev1 "k8s.io/api/core/v1"
simontype "github.com/hkust-adsl/kubernetes-scheduler-simulator/pkg/type"
gpushareutils "github.com/hkust-adsl/kubernetes-scheduler-simulator/pkg/type/open-gpu-share/utils"
"github.com/hkust-adsl/kubernetes-scheduler-simulator/pkg/utils"
)
const (
TagInitSchedule = "InitSchedule"
TagPostEviction = "PostEviction"
TagPostDeschedule = "PostDeschedule"
TagScheduleInflation = "ScheduleInflation"
TagDescheduleInflation = "DescheduleInflation"
)
// ClusterGpuFragReport Reports the Gpu Frag Amount of all nodes
func (sim *Simulator) ClusterGpuFragReport() {
nodeStatus := sim.GetClusterNodeStatus()
if len(nodeStatus) == 0 {
return
}
if sumRatio := utils.PodListRatioSum(sim.typicalPods); math.Abs(sumRatio-1) > 1e-3 {
log.Errorf("sim.ClusterGpuFragReport: (%.4f != 1.0): %v\n", sumRatio, sim.typicalPods)
}
sim.nodeResourceMap = utils.GetNodeResourceMap(nodeStatus)
clusterFragAmount := utils.NewFragAmount("cluster", make([]float64, len(utils.FragRatioDataMap)))
var clusterFragBellman float64
var clusterUsedNodes int // num. of used nodes
var clusterUsedGpus int // num. of used GPUs
var clusterUsedGpuMilli int64 // num. of used GPU in Milli
var clusterTotalGpus int // num. of total GPUs in the cluster
var nodeCnt int = 0
for _, ns := range nodeStatus {
if nodeRes, ok := sim.nodeResourceMap[ns.Node.Name]; ok {
clusterFragAmount.AddFragAmount(sim.NodeGpuFragAmount(nodeRes)) // easy to calculate. The regular Frag definition
clusterFragBellman += utils.NodeGpuFragBellman(nodeRes, sim.typicalPods, &sim.fragMemo, 1.0)
log.Debugf("[DEBUG][sim.ClusterGpuFragReport] calc [%d] node(%s) frag (%.2f): %s\n", nodeCnt, nodeRes.NodeName, clusterFragAmount.FragAmountSumExceptQ3(), clusterFragAmount.Repr())
// To calculate the allocation ratio of cluster in an online manner
clusterTotalGpus += nodeRes.GpuNumber
if nodeRes.GetFullyFreeGpuNum() < nodeRes.GpuNumber || nodeRes.MilliCpuLeft < nodeRes.MilliCpuCapacity { // the node is used
clusterUsedNodes += 1
clusterUsedGpus += nodeRes.GpuNumber // treat all GPUs on that node are "used"
clusterUsedGpuMilli += int64(nodeRes.GpuNumber*gpushareutils.MILLI) - nodeRes.GetTotalMilliGpuLeft()
}
}
nodeCnt += 1
}
var idleGpuMilli float64 // milli GPUs idle in the cluster
for _, v := range clusterFragAmount.Data {
idleGpuMilli += v
}
log.Debugf("[DEBUG][plugin.ClusterGpuFragReport] idleGpuMilli of %d nodes: %.2f\n", nodeCnt, idleGpuMilli)
fragGpuMilli := clusterFragAmount.FragAmountSumExceptQ3()
fragGpuRatio := 100 * fragGpuMilli / idleGpuMilli
q124GpuRatio := 100 * clusterFragAmount.FragAmountSumQ1Q2Q4() / idleGpuMilli
log.Infof("[Report]; Frag amount: %.2f; Frag ratio: %.2f%%; Q124 ratio: %.2f%%; (origin)\n", fragGpuMilli, fragGpuRatio, q124GpuRatio)
log.Infof("[Report]; Frag amount: %.2f; Frag ratio: %.2f%%; (bellman)\n", clusterFragBellman, 100*clusterFragBellman/idleGpuMilli)
//if clusterTotalGpus*gpushareutils.MILLI-int(idleGpuMilli) != int(clusterUsedGpuMilli) { // this prints unnecessary logs due to the int rounding error
// log.Errorf("totalGpuMilli (%d) - idleGpuMilli (%d) != usedGpuMilli (%d)", clusterTotalGpus*gpushareutils.MILLI, idleGpuMilli, clusterUsedGpuMilli)
//}
log.Infof("[Alloc]; Used nodes: %d; Used GPUs: %d; Used GPU Milli: %d; Total GPUs: %d; Arrived GPU Milli: %d\n", clusterUsedNodes, clusterUsedGpus, clusterUsedGpuMilli, clusterTotalGpus, sim.arrPodGpuMilli)
}
func (sim *Simulator) ReportFragBasedOnSkyline() {
nodeStatus := sim.GetClusterNodeStatus()
sim.nodeResourceMap = utils.GetNodeResourceMap(nodeStatus)
var clusterIdleMilliGpu int64 = 0
var clusterFragMilliGpu int64 = 0
for _, ns := range nodeStatus {
if nodeRes, ok := sim.nodeResourceMap[ns.Node.Name]; ok {
fragMilliGpu := utils.NodeGpuFragBasedOnSkyline(nodeRes, sim.skylinePods)
clusterFragMilliGpu += fragMilliGpu
idleMilliGpu := nodeRes.GetTotalMilliGpuLeft()
clusterIdleMilliGpu += idleMilliGpu
}
}
log.Infof("[Skyline] Frag GPU amount: %d, Idle GPU amount: %d, Frag GPU ratio: %.2f%%\n",
clusterFragMilliGpu, clusterIdleMilliGpu, 100.0*float64(clusterFragMilliGpu)/float64(clusterIdleMilliGpu))
}
func (sim *Simulator) ClusterAnalysis(tag string) (utils.FragAmount, []utils.ResourceSummary) {
nodeStatus := sim.GetClusterNodeStatus()
if len(nodeStatus) == 0 {
return utils.FragAmount{}, nil
}
sim.nodeResourceMap = utils.GetNodeResourceMap(nodeStatus)
ch := make(chan utils.FragAmount)
for _, ns := range nodeStatus {
go func(ns simontype.NodeStatus) {
var nodeFragAmount utils.FragAmount
if nodeRes, ok := sim.nodeResourceMap[ns.Node.Name]; ok {
nodeFragAmount = sim.NodeGpuFragAmount(nodeRes)
} else {
log.Errorf("nodeName %s not found in nodeResourceMap\n", ns.Node.Name)
}
ch <- nodeFragAmount
}(ns)
}
chCount := 0
data := make([]float64, len(utils.FragRatioDataMap))
clusterFragAmount := utils.NewFragAmount("cluster", data)
for nodeFragAmount := range ch {
if err := clusterFragAmount.AddFragAmount(nodeFragAmount); err != nil {
log.Errorf("[ClusterAnalysis] %s\n", err.Error())
}
log.Tracef("[%3d] Frag %s\n", chCount, nodeFragAmount.Repr())
chCount += 1
if chCount == len(nodeStatus) {
break
}
}
nodeAllocMap, err := utils.GetNodeAllocMap(nodeStatus)
if err != nil {
log.Errorf("[ClusterAnalysis] %s\n", err.Error())
}
log.Infoln()
log.Infof("========== Cluster Analysis Results (%s) ==========", tag)
resourceSummaries := utils.ReportNodeAllocationRate(nodeAllocMap)
var gpuFragSum float64
var FragRatioDataReverseMap = map[int]string{}
for k, v := range utils.FragRatioDataMap {
val := clusterFragAmount.Data[v]
gpuFragSum += val
FragRatioDataReverseMap[v] = k
}
for v := 0; v < len(utils.FragRatioDataMap); v++ {
k := FragRatioDataReverseMap[v]
val := clusterFragAmount.Data[v]
log.Infof("%-13s: %6.2f x 10^3 (%5.2f%%)\n", k, val/1000, 100*val/gpuFragSum)
}
log.Infoln("--------------------")
log.Infof("%-13s: %6.2f x 10^3 (100.0%%)\n", "idle_gpu_milli", gpuFragSum/1000)
val := clusterFragAmount.FragAmountSumExceptQ3()
log.Infof("%-13s: %6.2f x 10^3 (%5.2f%%)\n", "frag_gpu_milli", val/1000, 100*val/gpuFragSum)
log.Infoln("==============================================")
log.Infoln()
return clusterFragAmount, resourceSummaries
}
func (sim *Simulator) NodeGpuFragAmount(nodeRes simontype.NodeResource) utils.FragAmount {
if len(sim.typicalPods) <= 0 {
log.Errorf("Typical pods are not set.\n")
return utils.FragAmount{}
}
nodeResKey := nodeRes.Flatten("origin")
if fa, ok := sim.fragMemo.Load(nodeResKey); ok {
if fragData, ok2 := fa.([]float64); ok2 {
frag := utils.NewFragAmount(nodeRes.NodeName, fragData)
log.Debugf("[DEBUG][sim.NodeGpuFragAmount] load node(%s) frag (%.2f): %s\n", nodeRes.NodeName, frag.FragAmountSumExceptQ3(), frag.Repr())
return frag
}
} else {
frag := utils.NodeGpuShareFragAmount(nodeRes, sim.typicalPods)
sim.fragMemo.Store(nodeResKey, frag.Data)
log.Debugf("[DEBUG][sim.NodeGpuFragAmount] calc node(%s) frag (%.2f): %s\n", nodeRes.NodeName, frag.FragAmountSumExceptQ3(), frag.Repr())
return frag
}
log.Debugf("[DEBUG][sim.NodeGpuFragAmount] miss node(%s) frag: <NULL>\n", nodeRes.NodeName)
return utils.FragAmount{}
}
// RecordPodTotalResourceReq will record the total resource requests of all pods,
// and return the list: [total milli cpu request of all pods, total milli gpu request of all pods]
func (sim *Simulator) RecordPodTotalResourceReq(pods []*corev1.Pod) (int64, int64) {
// initialization
sim.podTotalMilliCpuReq = 0
sim.podTotalMilliGpuReq = 0
// count them all
for _, p := range pods {
podRes := utils.GetPodResource(p)
sim.podTotalMilliCpuReq += podRes.MilliCpu
sim.podTotalMilliGpuReq += podRes.TotalMilliGpu()
}
log.Infof("Total milli cpu request of all pods: %d, milli gpu request: %d\n",
sim.podTotalMilliCpuReq, sim.podTotalMilliGpuReq)
return sim.podTotalMilliCpuReq, sim.podTotalMilliGpuReq
}
// RecordNodeTotalResource will record the total resources of all nodes,
// and return the list: [total milli cpu of all nodes, total milli gpu of all nodes]
func (sim *Simulator) RecordNodeTotalResource(nodes []*corev1.Node) (int64, int64) {
// initialization
sim.nodeTotalMilliCpu = 0
sim.nodeTotalMilliGpu = 0
// count them all
for _, n := range nodes {
sim.nodeTotalMilliCpu += n.Status.Capacity.Cpu().MilliValue()
sim.nodeTotalMilliGpu += int64(gpushareutils.GetGpuMilliOfNode(n))
}
log.Infof("Total milli cpu of all nodes: %d, milli gpu: %d\n", sim.nodeTotalMilliCpu, sim.nodeTotalMilliGpu)
return sim.nodeTotalMilliCpu, sim.nodeTotalMilliGpu
}
func (sim *Simulator) SetTypicalPods() {
sim.typicalPods = utils.GetTypicalPods(sim.workloadPods, sim.customConfig.TypicalPodsConfig)
if sumRatio := utils.PodListRatioSum(sim.typicalPods); math.Abs(sumRatio-1) > 1e-3 {
log.Errorf("sim.SetTypicalPods: (%.4f != 1.0): %v\n", sumRatio, sim.typicalPods)
}
sim.fragMemo = sync.Map{}
}
func (sim *Simulator) SetSkylinePods() {
sim.skylinePods = utils.GetSkylinePods(sim.workloadPods)
}
func (sim *Simulator) NodeGpuFragAmountMap(nodeResourceMap map[string]simontype.NodeResource) map[string]utils.FragAmount {
nodeFragAmountMap := make(map[string]utils.FragAmount)
for nodeName, nodeRes := range nodeResourceMap {
nodeFragAmountMap[nodeName] = sim.NodeGpuFragAmount(nodeRes)
}
return nodeFragAmountMap
}