1-Computing Infrastructures PDF

Preview

Summary

Warning: TT: undefined function: 32 Warning: TT: undefined function: 32COMPUTING INFRASTRUCTURESIntroductionData centers are very important nowadays because they have made impressive transformations, and thanks to the connection with the Internet, they allow the creation of the cloud computing, a ne...

Description

COMPUTING INFRASTRUCTURES Introduction Data centers are very important nowadays because they have made impressive transformations, and thanks to the connection with the Internet, they allow the creation of the cloud computing, a new technology to operate and compute different types of operations and services. But how and what can we do when just a single personal computer is not enough? We all have a personal computer but there are some constraints in terms of computational power or memory. For example, if we consider deep learning techniques, in particular deep learning neural networks, they require a lot of competition, a lot of memory, lots of data and very often a single personal computer is not enough to solve these problems so we have to move our computation to a computer infrastructure able to run the schools in a very efficient way. Topics of the course But we have to say that cloud also has some constraints and limitations, so we will see how architectures are organized and we will also cover the basics of the data centres which are nowadays available. We will also understand and evaluate also the performance of a single computer ranging from a single computer going to the global infrastructure and we will also have a look all the dependability issues so all the techniques and analysis which allow us to evaluate the dependability of a system. The second part of the course will provide a quantitative and qualitative evaluation of systems in order to be able to evaluate, measure, predict and also ensure target behaviours. We will talk about the “queuing theory” which is related to what happens when you have a lot of jobs which have to be carried out and you have scars resources and so you might have long queue and delays. WHAT IS A COMPUTING INFRASTRUCTURE? A computing infrastructure is a technological infrastructure that provides hardware and software for computation to other systems and services. This definition is very important because at first a technological infrastructure is not just hardware, but it is hardware and software so computing infrastructure is not just the data centre itself, it is not just the server itself that we are using in order to perform some computation logical infrastructure. It is providing computation to other systems and services. This definition does not refer only to data centres because all the infrastructure providing the hardware and software can be considered a competing infrastructure. We might consider different types of computing infrastructure, ranging from a large scale data centres to all the computing infrastructure which can operate directly in the world or in the environment (pervasive computing infrastructures). Examples of these urban supercomputing infrastructures are: -

the edge computing systems the embedded personal computers the Internet of Things

These families are very different each other because, for example, IoT consists in tiny devices which are characterised by very strong constraints on computational memory and very often they are battery powered. But there are other systems that do not have problem of energy and they could also be endowed with an accelerator to run high performance computing algorithms. These families of technologies are characterised by very different performances and very different constraints on computation.

DATA CENTERS: A TECHNOLOGICAL PERSPECTIVE A data centre is a collection of servers; we have different types of servers: -

server for processing characterized by a given architecture server for the storage which are specifically designed for storing data server for communications: it is important to have a very efficient communication among all the other servers

A computing infrastructure comprises hardware and software and the servers are part of hardware. The software is made by virtual machines and containers. Virtual machines: there are different types of virtual machines. The goal of the virtual machine is to provide the full stack ranging from operating system to library and application. There are also virtual machine which just provide the libraries in the application.

A Container is an application which is packaged so it is built up with all the dependencies into a standardized unit which is used for software development and then deployment. Once we are configuring an application, we are developing the application with the final container and as soon as it is ready to be made into an operational life and then it is moved to the deployment.

Advantages of data centers: 1. Lower IT costs: the idea of data centre is to reduce the costs related to the information technology because instead of having lots of personal computers which have to be updated or configured, we are centralizing all the components and this allows us to reduce the cost.

2. High performance: data centers have a very efficient and effective architecture so we can increase the performance. 3. Instant software updates: we can move and update for example adding a patch to our operating system 4. “Unlimited” storage capacity: instead of having just one single hard disk, we can rely on multiple sets of storage servers so we can rely on a fairly unlimited storage capacity 5. Increased data reliability: we can rely on techniques that allow us to increase the reliability 6. Universal document access: if we think about very popular applications like Dropbox, we can access all our documents which are stored in our drop box directory from our personal computer or mobile phones. 7. Device independence: this guarantees some sort of device independence because the computation is carried out on the cloud and it is not related to the device we are using to access the cloud Disadvantages of data centers: 1. Require a constant internet connection: in order to access to data we need an Internet connection that must be steadily available because we want to connect to our data center at each time of the day 2. Do not work well with low-speed connections: when we are in an area where the wireless communication is not very powerful we have problem because we have problem in accessing our data and this is a disadvantage when all the data are on the cloud 3. Hardware Features might be limited: if we need a specific type of hardware accelerator for example GPU or TPU we can use it in the cloud but only the designer of data centre can load or insert these hardware accelerators on each data centre; otherwise we cannot use it 4. Privacy and security issues: often we have data related to our personal information 5. High power consumption: probably it's hard to believe the amount of energy these data centres can burn 6. Latency in making decision: if you are designing a security system which has to identify the presence of a person in a given dangerous area and you are using video cameras and you design an algorithm able to identify the person, but you are running this algorithm on the cloud, it can be a problem. We have to capture the image and move the message to the cloud. The image has to be decoded and the message processed by the deep learning neural network. If a person is recognised, we have to send the message back to the security system and then through the Internet and then an alarm must be rated. We have some time constraints so this approach is not the best one because we have a long latency in make the decision.

EDGE AND FOG COMPUTING SYSTEMS This is the reason why we are now considering the computing infrastructures that are operating in a pervasive way within our environment, like edge computing systems, embedded personal computers and embedded devices. The edge and fog computing architecture are very popular because they are reducing the distance between the real world (for example when we have IoT applications in the house or car) and the cloud. For example when we have problems in of latency in

making decision (if we move everything to the cloud), these problems can be managed by the edge and fog computing systems. The idea is to move the computation close to where data are generated and this is exactly what edge computing systems and for computing systems do: they are small servers or small cloud which operate close to where data are generated. There are different technologies that can be considered for edge and fog computing: technology ranges from an Intel Atom Processor, with a reduced amount of ram, going up to an Intel Xeon Processor, with lots of gigabytes for ram. So there is not unique technology. Advantages of edge and fog computing systems: 1. High computational capacity 2. Distributed computing: they can exchange data and computation through the offloading paradigm 3. Privacy and security: for example if we have a hospital and we want to carried out analysis and evaluation on the data which are related to the people which are working or the patients, we can use an edge computing system in order to carry out this computation without requiring the movement of all the data to the cloud which could be a problem in terms of privacy of the data 4. Reduced latency in making a decision: we can make a decision as soon as we acquire and process the data without the need to send everything to the cloud Disadvantages of edge and fog computing systems: 1. Require a power connection: these servers are powerful but they also requiring very high energy so they cannot be battery powered but they required a power line in order to provide energy 2. Require connection with the Cloud: in order to exchange information

EMBEDDED PERSONAL COMPUTERS Embedded personal computer are a sort of personal computer which is operating in a pervasive way. Examples are: -

Raspberry Pi 3B: it has 1GB of RAM and one Arm11 CPU Ordoid family Jetson TX2: it is an embedded personal computer endowed with a hardware accelerator Google coral: it is endowed with a TPU

Advantages of embedded personal computers: 1. Pervasive computing 2. High performance unit 3. Availability of development boards: you can create or develop the code and test embedded personal computer and then move it to the deployment 4. Programmed as a personal computer 5. Large community: that you can use in order to find updates algorithms and codes Disadvantages of embedded personal computers:

1. Pretty high power consumption: specifically for those embedded personal computers which are endowed with hardware accelerator (Jetson or Google Coral) 2. Some HW design has to done: for example connection with the sensors and actuators

INTERNET OF THINGS These technologies are very constrained in terms of computation and memory. Examples are Arduino or STM32. -

The CPU could be an ARM Cortex M3 or M4. The frequency which ranges from 30 to some hundreds of MHz The memory, in terms of RAM memory, is very limited going up to 320 as well as the FLASH memory which could go up to 2 megabytes.

Advantages of IoT: 1. 2. 3. 4. 5.

Highly pervasive Wireless connections Battery powered Low costs Sensing and actuating: they can be endowed with the ability to acquire data from the environment and provide commands to environment

Disadvantages of IoT: 1. Low computing ability 2. Constraints on energy, because being battery powered energy is considered for acquiring data or using actuators or transmitting information on communications 3. Constraints on memory (RAM/FLASH) 4. Difficulties in programming: there are some difficulties because most of these devices are not endowed with an operating system and they are considered embedded devices. IT PERSPECTIVE FOR COMPUTING INFRASTRUCTURES In our computing systems we have three different layers: -

-

Hardware: comprises all the physical resources of the system like storage, communication, input/output. It is the DATACENTER Software (environment): it is related to the virtual machine manage. Here we have all the programs and libraries that control the physical resources and provide tool in order to build application. Thanks to this layer we can build up an application and run it. It is the CLOUD Software (applications): it is the reason why a system exists....