CPI Creation Pre-Flight Checklist

checklist.md

• Machine properties
  • What predefined types are supported
  • Are custom machine types supported
  • Can arbitrary machine settings be provided at creation time
    • Are they available within the machine
    • Can they be updated after the machine has started
  • How many machines can be created by default
    • How can this limit be raised
• Is it possible to upload custom machine images
  • What image formats are supported (qcow2,raw,vmdk)
  • Are there any operating system restrictions
    • Are any software packages required or recommended
  • Can machine images be made outside of the provider
• What virtualization types are supported (hypervisor not packaging)
• What support exists for persistent disks
  • Is the lifecycle of a persistent disk independent of the machine it is attached to
  • What are the size constraints on an individual disk
  • How many disks can be created
  • Are there any restrictions on the number of disks which can be attached to a single machine
  • Is disk I/O configurable; what are the maximum and minimum throughput and is this per disk or is I/O shared among a pool of disks
• What support exists for ephemeral disks
  • Is it possible to create disks that have a lifecycle associated with a machine
    • Can emepheral disks be specified at machine creation time
    • Will the contents survive a machine power cycle
  • Are there limits on the number or size of ephemeral disks which can be attached to a machine
• General disk handling
  • Are disk attachment hints available outside the machine, do disk attachment hints from the IaaS correlate in any way with attachment hints inside the machine (e.g. UUID, device name)
  • Can disks be attached and detached at any time (e.g. does the machine have to be stopped)
  • Can a snapshot be created from a disk at any time (e.g. does the disk have to be detached from the machine)
• Software defined networking
  • Can multiple networks be created
  • Is creating a network with an arbitrary CIDR block supported
  • Can a network be devided into multiple subnets
  • Is it possible to assign a machine an arbitrary static ip
  • Is DHCP supported
  • Is DNS available
  • Is it possible to automatically assign public IP addresses
  • Is IPv6 supported
  • How many networks can be created
  • Is network performance customizable, are network throughput settings defined on a network, machine, or both
  • Can a specific static public address be associated with a machine
    • Does this address have a lifecycle independent of the machine
    • Can this address be assigned to another machine
• Firewall support
  • Can inbound and outbound traffic rules be specified
  • Are traffic rules applied to a machine, a network, or both
  • If firewall rules can be applied to a machine, can they be set at any time during the machine lifecycle
  • Can firewall rules be modified after creation
• Is it easy to program against
  • Is an SDK available in Ruby or Golang
    • Are they open source; what license is used
  • Is a RESTful API documented
• Are products besides compute and block storage available
  • Load balancing
  • Blob storage
    • Is a custom API used or is there an S3/Swift compatible API exposed
  • Relational database
• Availability zones
  • Are availability zones a first class concept
  • What resources can or cannot be moved accross availability zones
  • Is it possible to clone resources into different availability zones
• Responsiveness and reliability
  • Does creating a new machine take roughly one minute, five minutes, ten minutes or more
  • Does the speed at which machines are created depend on how many machines have already been created
  • Are machines created on shared hardware
    • Is it possible to create machines on dedicated hardware
  • For hosted solutions, can an SLA be purchased
  • What is the advertised relaibility of the solution
• Pricing considerations
  • Can machines be requested at a "market" price (e.g AWS Spot Instances)
  • Can machines be requested at a cheaper price with less reliable service (e.g. GCE Preemptable VMs or AWS Burst Instances)
• Hosted Regions
  • Are geographically distributed service regions available
    • Are mutliple access credentials required to address multiple regions
    • Are any regions isolated or special (e.g. China)
• On premise installation
  • How many machines are required for a standard installation of the solution
  • Is there a single machine development version available
  • Are there hosting providers for the solution

scratch.md

• Are GCE concepts/naming similar to AWS's?

• How's GCE's responsiveness?

• How's GCE's reliability?

  • API reliability (i.e. starting a VM fails 30% of the time)
  • Infrastructure reliability (i.e. after 2 days, I was unable to SSH into my VM)

• Security features. How does it measure up to AWS?

• Anything else we need to know?

• For hosted solutions are there any unique regions (e.g. China) how do they compare with "standard" regions

• For on premise solutions how hard is it to deploy (number of machines, software available as binaries)

• For on premise solutions, is there a local development version and how full featured is it