Appunti Computing Infrastructures

Types of Cooling Systems and Data Centers

A three-loop system is the most expensive to construct and has moderately complex controls, but offers contaminant protection and good efficiency.

Liquid cooling is more "dangerous" and difficult to maintain, but it offers a good solution for very complex systems. Near future will lead to In-Rack Cooler where an air-to-water heat exchanger is located at its back, so the hot air exiting the servers immediately flows over coils cooled by water decreasing the path or we can directly cool server components using cold plates using a liquid loop.

Cooling usually requires about half the energy required by the IT equipment.

Container-based data centers go one step beyond in-row cooling by placing the server racks inside a container that can be moved, typically it is provided of all its dependencies.

Tier 1: Nothing

Tier 2: Redundant Site Infrastructure

Tier 3: Redundant Site Infrastructure and IT equipment dual powered

Tier 4: Redundant Site Infrastructure, IT equipment and cooling system dual powered

Cloud Computing

Cloud computing allows us to better fit the computational capability demand of our system. We can dismiss or purchase new servers according to the demand, reducing the cost and improving performance.

Cloud is implemented thanks to the partition of hardware resources shared among multiple virtual machines. The virtual machine monitor (VMM) governs the access to the physical resources.

With virtualization we have that our system is:

• Hardware Independent: software now are not linked to the hardware
• High Flexible

Virtualization leads to cloud computing and Server consolidation.

12.1 Server Consolidation

Thanks to virtualization we can run different OS on the same hardware. This leads to higher hardware utilization but also lets us achieve application independency from the hardware. Migrating to virtual machines allow us to change hardware without even stop the services and balance the workload among different machine as if we are still using a single one, looking at the group of servers.

as a single VM infrastructure
We can also continue to use legacy software using out-of-date OSs
12.2 Cloud Computing
Cloud computing is a model for enabling on-demand computing systems. With it we have network access to a shared pool of configurable computing resources that can be rapidly provisioned and released with minimal management effort.
The more we leave the management of the system to external enterprise, the more we are exploiting cloud computing. We have different levels that can be grouped as follow:
- Cloud Application: SaaS (software): users access the services provided by this layer through web-portals and are sometimes required to pay fee to use them
- Cloud Software Environment PaaS (platform): users are application developers. Provider supply a program-language-level environment with a well-defined API to simplify interaction between environments and apps
- Cloud Software Infrastructure: IaaS (infrastructure), DaaS (storage), CaaS (communication). Provide resources to the

higher-level layer.- IaaS allow us super-user access to Vm for fine granularity setting but decrease performance and is not able to provide guarantees abous service level agreement

DaaS allow user to store their data at remote disks and access anytime from anywhere

Caas: Communication is outed to external company that must guarantee a QoS

Types of clouds:

- Public Clouds: large scale infrastructure available on a rental basis. Large scalability, accessed through the network and has to guarantee SLas

- Private Cloud: Internally managed Data Center. The Organization sets up a virtual environment on its own internal services. The pro is that you have the total control over the infrastructure, but you have a tradeoff in term of scalability and flexibility

- Community Clouds: A single cloud managed by several federated organizations. Useful to share costs, have good control over the infrastructure and a reasonable level of scalability. A more complex accounting system is needed.

- Hybrid Cloud:

Combination of previous clouds13. VirtualizationHardware resources are partitioned and shared among multiple virtual machines. In computer architecture, the set of instructions that a program can use is structured at different levels.We usually program the software part (5-3) and exploit the hardware (2-1).Level 2 ISA (7+8) marks the division between hardware and software.- User ISA: aspect of the ISA that are visible to an application program, when application interacts with the HS, hardware is used- System ISA: aspects visible to supervisor software which is responsible for managing hardware.When the OS interacts with the HW, system ISA is usedLevel 3 ABI (7+3) - Application binary interface provides a program with the hardware resources and the services available in the system. A program running at ABI level directly uses the user instruction at CPU, but calls O.S. procedures to access the hardware resources.- User ISA- System Calls: call that allow programs to interact with shared

hardware resources indirectly by OS

One machine level can only run instructions that were meant for it. A Virtual Machine (VM) is a logical abstraction able to provide a virtualization execution environment. it tasks are:

• To map virtual resources or states to corresponding physical ones
• to use physical machines instructions\calls to execute virtual ones

Emulation: Different ABI/ISA. It allows software to run in an environment different from the architecture originally intended. It reads all the bytes included in the memory of the system it is going to reproduce. Provide Both ISA and ABI, useful when you need to run a program with different ISA\ABi from hosts

2 types of VM:

1. Process VMs: the runtime software supports the level 0-3 of the architecture. it is able to support an individual process. The virtualization is placed at ABI interface
2. Multi Programmed system: Same ABI/OS. VM formed by OS call interface + User IA. Each user process is given the illusion of having a complete machine

itself and its own address space. - High-Level Language VM: the goal is to have an isolated execution environment for each application (like java virtual machine) - System VMs: supports the 0-2 levels of the architectures. Provide a complete system environment that can support an operating system. It is placed at ISA interface level. This VM can work directly on hardware or running other OSs. - Same ISA: it can intercept the guest OS's interaction with hardware resources. it is the most efficient VM architecture: HW executes the same instruction of VMs. 2 different Os on same HSs - Hosted VM: virtualizing software is on top of an existing host operating system - Whole-system VMs: virtualize all software: ISAs are different so both application and OS code require emulation. Usually implemented the VMM and guest software on top of a host OS 13.1 Implementation In hardware level virtualization the virtualization layer is placed between hardware and OS. The interface seen by the OS and

In Application level virtualization: the virtualization layer is placed between the OS and some application. Application run in their environment independently from the OS.

In System Level virtualization the virtualization layer provides the interface of a physical machine to a secondary OS and a set of applications running in it. Placed between the system OS and other OS.

13.1 Virtual Machine managers (System VMs- Same ISA)

The virtual machine manager is an application that: manages the virtual machines, mediates access to hardware resources and intercepts and handles any instruction issued by the VM. This type of virtualization typically runs virtual machine whose systems are of the same type of the host machine.

3 terms to identify the same thing: Virtual Machine Manager, VM monitor, Hypervisor.

2 kind of hypervisor:

• Type 1: take direct control of the hardware. Differentiation: Monolithic - Devices drivers run with the hypervisor. Better performance.

but can run only on hardware for which the hypervisor has driver

MicroKernel: Device drivers within a service VM. Smaller hypervisor, driver ecosystem of an existing OS. Can use 3rd party driver

Type 2: Reside within a host operating system. At least 2 OS running on the same hardware, More flexible in terms of underlying hardware and simpler to manage and configure. You have to manage conflict among OSs and host OS can consume non-negligible set of resources

13.2 Virtualization Technique

we have different way to implements (System Level Same ISA) virtualization

Paravirtualization: Guest OS and VMM collaborates:

• VMM present to VMs an interface similar but not identical to that of the underlying hardware
• reduce guest's execution of task too expensive from the virtualized environments

Thanks to it we have simpler VMM and High performance, but we have to modify the guest OS

Full Virtualization: Provides a complete simulation of the underlying hardware. Some protected instructions are

trapped and handled by hypervisors. Pros to run unmodified OS, but limitation on architectures and performance

13.3 Containers

Containers are pre-configured packages, with all you need to execute code in the target machine. The main advantage of containers is that their behavior is predictable, repeatable, and immutable.

The main difference is that containers share the host system kernel with other containers. Containers are flexible, light, interchangeable, and portable. They also scale well and are stackable. But they impose a modular application development where the modules are independent and uncoupled.

Useful to build a multi-user platform as a Service, they also provide lightweight stand-alone sandbox environments for developing and Software as a Service Applications

Main Feature:

• Encapsulation: allow to save the state of a running VM into a single file
• Partitioning: allow to run multiple O.S. on the same physical machine
• HW-Independent: allow to migrate VM on other hardware

Isolation: allow the user to share the same machine with independent data

14. Performance modeling

Computer performance in terms of throughput, response time and availability. To evaluate system quality we have different approaches:

• Intuition and Trend Extrapolation: Rapid and flexible, but not accurate and need people know-how.
• Experimental Evaluation: Excellent accuracy, but laborious and inflexible, typically costly.
• Model Based Approach: often the only reasonable method. Cheap and Accurate.

Model based techniques are:

• Analytical and Numerical: are based on the application of mathematical techniques. Very efficient and precise, but limited in few cases.
• Simulation: are based on the reproduction of traces of the model. Most general but less accurate. Can hide rare events.
• Hybrid Technique

14.1 Queueing network

Queueing network modeling is a particular approach to computer system modeling in which a computer system is represented as