Investigate Autoscaling

[jeffmccune]

Discovery

See: Autoscaling groups of instances

Decisions

• Use Cloud Monitoring metrics
• Use Per-group metrics (beta)
• Compute the group's bandwidth utilization.
• Target 80% utilization

Metrics of use:

The capacity and current_utilization metrics of the autoscaler might be useful.

capacity - Utilization target multiplied by number of serving VMs. Sampled every 60 seconds. After sampling, data is not visible for up to 180 seconds.

current_utilization - The sum of the utilization of a specified metric for all serving VMs. Sampled every 60 seconds. After sampling, data is not visible for up to 180 seconds.

instance/network/sent_bytes_count - Delta count of bytes sent over the network. Sampled every 60 seconds. After sampling, data is not visible for up to 240 seconds.

l3/internal/ingress_bytes_count - The number of bytes sent from client to ILB backend (for TCP flows it's counting bytes on application stream only). Sampled every 60 seconds. After sampling, data is not visible for up to 150 seconds. client_network: Network of the client instance in ILB flow. client_subnetwork: Subnetwork of the client instance in ILB flow. client_zone: Zone of the client instance in ILB flow.

[jeffmccune]

Experiment:

Time series per instance compute.googleapis.com/instance/network/sent_bytes_count

Utilization target: 250000000 bytes/sec (2Gbps) * 80% = 200,000,000 bytes/sec