Clouds

Five essential characteristics:

1. On-demand self-service: users do not need human help to add service.
2. Broad network access: resources are available over the network via standard mechanisms (e.g., http).
3. Resource pooling: users acquire resources from a common pool, and have no idea where "their" resources live.
4. Rapid elasticity: fast an automatic scaling has loads change.
5. Measured service: the services are metered much like electricity usage.

• Infrastructure as a service (IAAS): Offers direct access to a VM user can use as they see fit.
• Platform as a service (PAAS): Includes things such as web servers and databases.
• Old worries gone: backup, power, cooling, scalability, upgrading, etc.
• New worries arrive: is data secure? What laws apply? Can a co-hosted competitor snoop?
• Vendor lock-in: once you choose a cloud provider, can you ever (practically) leave?

How to migrate to faster hardware, off of broken hardware, etc.?

We could shut down all virtual machines and re-start them on new hardware: significant pause!

We could pause them, transfer their memory, and resume: still a noticable pause.

Or, using pre-copy memory migration, we can copy pages first, then do a live migration. If no one notices, it was a seamless live migration!

• Captures the machine at various times, so we can rollback if we need to.
• Use copy on write to save time: only copy what has changed.
• Unsolvable: what to do about remote machines that were interacting with ours? A: Nothing can be done (in general).

1. Elasticity refers to
1. the ability of paging to make memory appear larger than it is
2. the change in demand for a good in response to price
3. the ability of cloud services to add and drop resources swiftly
4. the way cloud providers stretch the truth in their advertising
2. In a seamless live migration
1. no users notice there was a migration
2. VMs only need to be shut down for a few minutes
3. the data center never closed during the migration
4. no lawsuits were filed against the cloud provider
3. An insurmountable problem for VM checkpointing is
1. what to do about large disks
2. what to do if the machine crashes
3. how to handle security
4. restoring remote interactions at the time of the checkpoint.