Ankur Upadhyaya | 71795a3 | 2023-04-17 10:05:47 +0530 | [diff] [blame] | 1 | .. |
| 2 | SPDX-FileCopyrightText: 2023-present Intel Corporation |
| 3 | SPDX-FileCopyrightText: © 2020 Open Networking Foundation <support@opennetworking.org> |
| 4 | SPDX-License-Identifier: Apache-2.0 |
| 5 | |
| 6 | .. _auto-scaling-5g-nfs: |
| 7 | |
| 8 | Auto scaling 5G NFs |
| 9 | =================== |
| 10 | |
| 11 | Autoscaling cloud native network functions is a critical capability for modern cloud infrastructure. |
| 12 | It enables dynamic scaling of network functions to handle increased traffic or workload demands, |
| 13 | ensuring optimal performance and cost-effectiveness. |
| 14 | |
| 15 | Kubernetes Event-driven Autoscaling (KEDA) is an open-source tool that makes it easier to implement |
| 16 | autoscaling for cloud-native network functions(https://github.com/kedacore/keda). |
| 17 | |
| 18 | When the network functions receive more traffic or workload, KEDA automatically scales up the pods to |
| 19 | handle the increased demand. When the demand decreases, KEDA scales down the pods to save resources and |
| 20 | minimize costs. |
| 21 | |
| 22 | We can enable autoscaling based on memory usage, CPU usage, and custom metrics. |
| 23 | |
| 24 | Illustration |
| 25 | ------------ |
| 26 | |
| 27 | Let's explore the steps on how to set up autoscaling in AIAB. |
| 28 | |
| 29 | In this example, we are setting up KEDA to scale the smf pod up for every 50 N4 messages received by SMF |
| 30 | |
| 31 | |
| 32 | Run the following steps in aether-in-a-box folder: |
| 33 | |
| 34 | * Create aiab.diff file as described below |
| 35 | * patch < aiab.diff |
| 36 | * Create resources/keda.yaml as described below |
| 37 | * Create resources/5g-monitoring/smf-monitor.yaml as described below |
| 38 | * Create autoscale.yaml as described below |
| 39 | * make 5g-core |
| 40 | * make monitoring-5g |
| 41 | * make autoscale-aiab |
| 42 | * kubectl get hpa -n omec : To view the horizontal pod scaler. |
| 43 | * kubectl get pods -n omec | grep smf : To view the scaled pods. |
| 44 | |
| 45 | Create file aiab.diff with following content |
| 46 | |
| 47 | .. code-block:: |
| 48 | |
| 49 | diff --git a/Makefile b/Makefile |
| 50 | index bd54a7a..df85e0a 100644 |
| 51 | --- a/Makefile |
| 52 | +++ b/Makefile |
| 53 | @@ -26,9 +26,10 @@ GET_HELM = get_helm.sh |
| 54 | KUBESPRAY_VERSION ?= release-2.17 |
| 55 | DOCKER_VERSION ?= '20.10' |
| 56 | HELM_VERSION ?= v3.10.3 |
| 57 | -KUBECTL_VERSION ?= v1.23.15 |
| 58 | +KUBECTL_VERSION ?= v1.24.11 |
| 59 | |
| 60 | -RKE2_K8S_VERSION ?= v1.23.15+rke2r1 |
| 61 | +RKE2_K8S_VERSION ?= v1.24.11+rke2r1 |
| 62 | +#RKE2_K8S_VERSION ?= v1.23.15+rke2r1 |
| 63 | K8S_VERSION ?= v1.21.6 |
| 64 | |
| 65 | OAISIM_UE_IMAGE ?= andybavier/lte-uesoftmodem:1.1.0-$(shell uname -r) |
| 66 | @@ -65,6 +66,8 @@ ROUTER_HOST_NETCONF := /etc/systemd/network/10-aiab-access.netdev /etc/systemd |
| 67 | UE_NAT_CONF := /etc/systemd/system/aiab-ue-nat.service |
| 68 | |
| 69 | # monitoring |
| 70 | +AUTOSCALE_CHART := kedacore/keda |
| 71 | +AUTOSCALE_VALUES ?= $(MAKEDIR)/autoscale.yaml |
| 72 | RANCHER_MONITORING_CRD_CHART := rancher/rancher-monitoring-crd |
| 73 | RANCHER_MONITORING_CHART := rancher/rancher-monitoring |
| 74 | MONITORING_VALUES ?= $(MAKEDIR)/monitoring.yaml |
| 75 | @@ -675,6 +678,26 @@ test: | 4g-core $(M)/oaisim |
| 76 | fi |
| 77 | @grep -q "Simulation Result: PASS\|Profile Status: PASS" /tmp/gnbsim.out |
| 78 | |
| 79 | +autoscale: $(M)/autoscale |
| 80 | +$(M)/autoscale: $(M)/helm-ready |
| 81 | + helm repo add kedacore https://kedacore.github.io/charts |
| 82 | + helm upgrade --install --wait $(HELM_GLOBAL_ARGS) \ |
| 83 | + --namespace=autoscale \ |
| 84 | + --create-namespace \ |
| 85 | + --values=$(AUTOSCALE_VALUES) \ |
| 86 | + keda-aiab \ |
| 87 | + $(AUTOSCALE_CHART) |
| 88 | + touch $(M)/autoscale |
| 89 | + |
| 90 | +autoscale-aiab: $(M)/autoscale |
| 91 | + kubectl apply -f resources/keda.yaml |
| 92 | + |
| 93 | +autoscale-clean: |
| 94 | + kubectl delete -f resources/keda.yaml |
| 95 | + helm -n autoscale delete keda-aiab || true |
| 96 | + kubectl delete namespace autoscale || true |
| 97 | + rm $(M)/autoscale |
| 98 | + |
| 99 | reset-test: | oaisim-clean omec-clean router-clean |
| 100 | @cd $(M); rm -f omec oaisim 5g-core |
| 101 | |
| 102 | diff --git a/resources/5g-monitoring/kustomization.yaml b/resources/5g-monitoring/kustomization.yaml |
| 103 | index 96bc72b..0b757e9 100644 |
| 104 | --- a/resources/5g-monitoring/kustomization.yaml |
| 105 | +++ b/resources/5g-monitoring/kustomization.yaml |
| 106 | @@ -5,6 +5,7 @@ |
| 107 | resources: |
| 108 | - ./metricfunc-monitor.yaml |
| 109 | - ./upf-monitor.yaml |
| 110 | + - ./smf-monitor.yaml |
| 111 | |
| 112 | configMapGenerator: |
| 113 | - name: grafana-ops-dashboards |
| 114 | |
| 115 | |
| 116 | Create a file resources/keda.yaml with the following content |
| 117 | |
| 118 | .. code-block:: |
| 119 | |
| 120 | --- |
| 121 | apiVersion: keda.sh/v1alpha1 |
| 122 | kind: ScaledObject |
| 123 | metadata: |
| 124 | name: smf-scale |
| 125 | namespace: omec |
| 126 | spec: |
| 127 | scaleTargetRef: |
| 128 | kind: Deployment |
| 129 | name: smf |
| 130 | minReplicaCount: 1 |
| 131 | maxReplicaCount: 5 |
| 132 | cooldownPeriod: 30 |
| 133 | pollingInterval: 1 |
| 134 | triggers: |
| 135 | - type: prometheus |
| 136 | metadata: |
| 137 | serverAddress: http://rancher-monitoring-prometheus.cattle-monitoring-system.svc:9090 |
| 138 | metricName: n4_messages_total |
| 139 | query: | |
| 140 | sum(n4_messages_total{job="smf"}) |
| 141 | threshold: "50" |
| 142 | |
| 143 | Create file resources/5g-monitoring/smf-monitor.yaml with following content |
| 144 | |
| 145 | .. code-block:: |
| 146 | |
| 147 | apiVersion: monitoring.coreos.com/v1 |
| 148 | kind: ServiceMonitor |
| 149 | metadata: |
| 150 | name: smf |
| 151 | namespace: omec |
| 152 | spec: |
| 153 | endpoints: |
| 154 | - path: /metrics |
| 155 | port: prometheus-exporter |
| 156 | namespaceSelector: |
| 157 | matchNames: |
| 158 | - omec |
| 159 | selector: |
| 160 | matchLabels: |
| 161 | app: smf |
| 162 | |
| 163 | Add an empty autoscale.yaml in aiab folder. This file can be used to add override values for keda helm chart. |
| 164 | |
| 165 | .. code-block:: |
| 166 | |
| 167 | touch autoscale.yaml |
| 168 | |