OSM Autoscaling

This is a new feature available since Release FIVE, which allows to automatically scale VNFs with a VDU granularity and based on any available metric.

Reference diagram

The following diagram summarizes the feature:

• Scaling descriptors can be included and be tied to automatic reaction to VIM/VNF metric thresholds.
• Supported metrics are both VIM and VNF metrics.
• An internal alarm manager has been added to MON, so that both VIM and VNF metrics can also trigger threshold-violation alarms and scaling actions.

Scaling Descriptor

The scaling descriptor is part of a VNFD. Like the example below shows, it mainly specifies:

• An existing metric to be monitored, which should be pre-defined in the monitoring-param list (vnf-monitoring-param-ref)
• The VDU to be scaled (vdu-id-ref) and the amount of instances to scale per event (count)
• The thresholds to monitor (scale-in/out-threshold)
• The minimum and maximum amount of scaled instances to produce.
• The minimum time it should pass between scaling operations (cooldown-time)

scaling-group-descriptor:
-   name: "cpu_autoscaling_descriptor"
    min-instance-count: 0
    max-instance-count: 10
    scaling-policy:
    -   name: "cpu_scaling_policy"
        scaling-type: "automatic"
        cooldown-time: 120
        scaling-criteria:
        -   name: "cpu_autoscaling_criteria"
            scale-in-threshold: 20
            scale-in-relational-operation: "LT"
            scale-out-threshold: 80
            scale-out-relational-operation: "GT"
            vnf-monitoring-param-ref: "vnf01_cpu_util"
    vdu:
    -   count: 1
        vdu-id-ref: vdu01

Example

This will launch a Network Service formed by an HAProxy load balancer and an (autoscalable) Apache web server Make sure:

1. Your VIM has an accesible 'public' network and a management network (in this case called "PUBLIC" and "vnf-mgmt")
2. Your VIM has the 'haproxy_ubuntu' and 'apache_ubuntu' images located here
3. You run the following command to match your VIM metrics telemetry system's granularity, if different than 300s (recommended for this example is 60s or Gnocchi's "high archive-policy"):

docker service update --env-add OS_DEFAULT_GRANULARITY=60 osm_mon

Get the descriptors:

wget https://osm-download.etsi.org/ftp/osm-4.0-four/4th-hackfest/packages/webserver_vimmetric_autoscale_nsd.tar.gz
wget https://osm-download.etsi.org/ftp/osm-4.0-four/4th-hackfest/packages/webserver_vimmetric_autoscale_vnfd.tar.gz

Onboard them

osm vnfd-create webserver_vimmetric_autoscale_vnfd.tar.gz
osm nsd-create webserver_vimmetric_autoscale_nsd.tar.gz

Launch the NS

osm ns-create --ns_name web01 --nsd_name webserver_vimmetric_autoscale_ns --vim_account <VIM_ACCOUNT_NAME>|<VIM_ACCOUNT_ID>
osm ns-list
osm ns-show web01

Testing

1. To ensure the NS is working, visit the Load balancer's IP at the public network using a browser, the page should show an OSM logo and active VDUs
2. To check metrics at Prometheus, visit http://[OSM_IP]:9091 and look for osm_cpu_utilization and osm_average_memory_utilization (initial values could take some some minutes depending on your Telemetry system's granularity)
3. To check metrics at Grafana, just install the OSM preconfigured version (./install_osm.sh -o pm_stack) and visit http://[OSM_IP]:3000 (admin / admin), you will find a sample dashboard (the two top charts correspond to this example)
4. To increase CPU in this example to autoscale the web server, install Apache Bench in a client within reach (could be the OSM Host) and run it towards test.php

sudo apt install apache2-utils
ab -n 5000000 -c 2 http://[load-balancer-ip]/test.php
# This will stress CPU to 100% and trigger a scale-out operation in POL.
# In this test, scaling will usually go up to 3 web servers before HAProxy spreads to load to reach a normal CPU level (w/ 60s granularity, 180s cooldown)

Any of the VMs can be accessed through SSH to further monitor (with htop, for example), and there is an HAProxy UI at port http://[HAProxy_IP]:32700 (all credentials are osm / osm2018)