For the first Cloud Computing assignment, I wrote up the five exercises from Chapter 1 of the textbook.

1. Definition of Cloud Computing

Renting computing resources whenever you need them. It is utility computing: you pay for what you use, like electricity, gas or water. It started with the utility computing John McCarthy described in 1961.

Related keywords

• On-Demand Service, Flexibility, resource reuse
• Virtualization, Data Center
• IaaS, PaaS, SaaS
• Private, public, hybrid cloud
• Amazon EC2, Google, Azure

2. Ways to Increase Computing Performance and Capacity

The first thing that comes to mind is upgrading parts like the CPU and memory. But there is a limit to how far a single computer can grow.

So there are two directions.

• Clustering: tie several computers together and use them for one purpose. Combining machines to raise performance is built on orchestration.
• Resource partitioning: split one computer's resources and use them for several purposes. This is built on virtualization.

3. Three Categories of Computer Virtualization

Virtualization is divided into CPU virtualization, memory virtualization and I/O virtualization.

Virtualization Approaches

Divided by whether the guest OS is modified.

• Full virtualization: the guest OS is not modified.
• Para virtualization: the guest OS is modified.

Hypervisor

The software that implements virtualization. There are two kinds depending on where it is installed.

• Type 1 (native): the hypervisor is installed directly on the hardware, and virtual machines run on top of it.
• Type 2 (hosted): hardware → host OS → hypervisor → guest OS, in that order.

How Each Category Is Virtualized

CPU virtualization

• The hypervisor handles privileged instructions executed by the guest OS on its behalf.
• If the guest is not modified, the hypervisor traps the instructions and emulates them.
• If the guest is modified, those instructions are replaced with hypercalls that request the hypervisor directly.

Memory virtualization

• Address translation gains one more level: guest virtual address → guest physical address → real physical address.
• The hypervisor manages this translation with shadow page tables. Physical addresses inside a VM are fake, so a separate table maps guest virtual addresses straight to real addresses.
• Alternatively, a modified guest asks the hypervisor when it changes its page tables.

I/O virtualization

• Multiple VMs share physical devices such as disks and network cards.
• The hypervisor either emulates the device in software, or a virtualization-specific driver (virtio, etc.) is installed in the guest to communicate directly.

4. How HPC, HTC, Grid and Cloud Connect

The goal of HPC is to finish a single computation quickly. Supercomputers fall into this category.

If the goal is throughput, you use HTC. Since many jobs have to be processed, the key is making the most of idle resources.

Grid computing and cloud computing belong to HTC. Both play the same role of using idle resources, but they are implemented differently.

• A grid ties together idle computers scattered around the world. So it is built on orchestration.
• A cloud uses a hypervisor to split one computer's resources into virtual machines. That is virtualization.

5. Three Key Issues in Data Center Operation and the Cloud

All three issues can be solved with virtualization. It is thanks to the properties of virtual machines: resources can be split finely, VMs are isolated from each other, and a whole VM can be moved as a single file.