Cloud Information Systems Flashcards

Question

Instance Storage

Answer 1

(disco físico directamente conectado al servidor donde corre tu instancia EC2.) No es realmente persistente: Si apagas o pierdes la instancia (por ejemplo, por fallo de hardware), los datos se borran. Sobrevive a fallos de energía (no se borra si la instancia se reinicia),

Answer 2

Es un disco virtual que puedes conectar a tus instancias EC2. Funciona como un disco duro externo, pero en la nube. -virtual disk -block device, not a file system -used as root volume for operating system -EBS volume should usually only be attached by one VM at any point in time, but can attach multiple volumes to one instance -can be replicated across multiple servers in one AZ -EBS Variants: io1, io2, st1, sc1

Answer 3

Es un servicio de almacenamiento en la nube donde puedes guardar archivos (fotos, documentos, videos, backups, etc.) de forma segura, escalable y económica. redundant storage across AZs in one region (buckets viven en la region pero se replican en todos los AZs) sometimes called object storage Terminology: “object” = file “prefix” = file path “bucket” = named collection of files in a region bucket lives in region, not AZ http(s) API: GET, PUT, LIST, DELETE (para acceder a los archivos) bucket can be internet public or private

Answer 4

-Keep everything in main memory (RAM) -Amazon Elastic File System (EFS): network file system (compartido por red) -DynamoDB: distributed key/value store -Relational Database Service (RDS): relational OLTP (online transaction processing) database system

Answer 5

Total energy consumption of the data center / Total energy consumption of the IT (servidores, redes, storage) Optimal PUE = 1.0 Legal Requirement: PUE <= 1.2 Total Energy Consumption (DC) = IT Equipment + HVAC + Lighting + Power Losses + Infrastructure + Miscellaneous (CCTV) Total IT Consumption = Servidores + Almacenamiento + Networking (swithces, router, firewall, load balancer) + Periféricos IT (backup systems)

Answer 6

-Increase number of AC units -Server fan replacement -Location in countries with low temperature (Finland)

Answer 7

Scale up: vertical scaling (bigger machines) -Enables high bandwidth and low latency communication -May not be enough Scale out; horizontal scaling (more machines) -Cloud enables elasticity: add/remove machines when workload changes -Enables fault tolerance through redundancy

Answer 8

More nodes -> higher the possibility that one of them fails network bandwidth and latency often becomes the bottleneck (en sistemas distribuidos esperar duele mas que no poder enviar mucho) New failure modes : network partitions (some nodes cannot talk to each other anymore), transient failures (fallos temporales), clock skew and drift (un nodo puede pensar que algo paso antes y otro puede pensar que algo paso despues)

Answer 9

Tomar una aplicación que ya existe en servidores propios (on-premise) y moverla a la nube casi sin cambiar nada. (copiar y pegar tu sistema actual pero en la nube) -Downsides of public cloud without benefits *pagas los caro de la nube pero sin recibir beneficios*

Answer 10

*en vez de una sola database, se dividen los datos en partes y esas partes se reparten en diferentes servidores* -multi-tenant service where each tenant has a small, independent database

Answer 11

*Separar las máquinas que calculan (compute) de las que guardan los datos (storage).* * scale compute and storage independently * separate billing for compute and storage

Answer 12

Common pattern control plane (brain): coordination, scheduling, monitoring, cost counting, etc. data plane (muscle): does the actual work

Answer 13

* each running in its own VM(s)/container(s) Each microservice: should be independent and loosely coupled (changes should not affect others) * should be scaled independently

Answer 14

Independent functionality Different scaling needs DIfferent technology stack

Answer 15

Advantages: Decoupled architecture All components scale independently Easy to add another detector Observed Problems: scalability bottlenecks due to per-customer account limits

Answer 16

Parallelism Consolidation More powerful machines Faster Network

Answer 17

Creation and deployment of new virtual resources Software update and upgrade (OS, applications, libraries) Administration of security policies (firewalls, secrets)

Answer 18

Treat infrastructure like software (version control, code review, testing) Immutable Infrastructure: Do not patch running servers; replace them with new, updated images.

Answer 19

It is usually not a good idea to manually configure and deploy a large number of heterogeneous containers Orchestration software automates this, and provides additional features like * dynamic scaling of services * coordination across multiple servers * resilience and automatic recovery

Answer 20

-Service naming and discovery In Kubernetes each service can be given a name, either * A domain name * An IP address -Load balancing Kubernetes uses load balancer software to divide requests among the instances of the service -Storage orchestration When the container is launched, Kubernetes connects each attachment point to external storage -Optimized container placement Assigns containers to nodes to optimize the use of the nodes (bin packing) -Automated Initiation and Recovery To handle automated failure detection and replacement, Kubernetes, * Continually probes each container with a user-defined health check * Terminates any container that stops responding -Mangement of configurations and Secrets Kubernetes separates * Service configuration and management information * Container images used to provide the service The reason: separation allows the owner of a service to change configuration and management policies that affect for deployment without rebuilding the container images used in the service -Automated Rollouts And Rollbacks -What Kubernetes does not do: Focus on a specific type of application or have application-specific optimizations Manage source code or build containers (e.g., assumes Docker handles it) Supply event-passing middleware Have a built-in facility to collect, log, or otherwise report measurements or events

Answer 21

* Cluster: set of nodes that run containerized applications * Node: physical or virtual machine in the cluster * Pod: smallest deployable unit Kubernetes Pods -All containers for a pod run on the same node -All containers in a pod share an IP address

Answer 22

+ Increase overall efficiency of operations + Coordinate computational, communication, and storage resource management + Avoid human errors in configuration and operation (= cost savings) − System-wide (cascading) failures − Risk of run-away resource use − Increased security attack surface − Complexity, overlapping functionality

Answer 23

Cada acceso a S3 es lento comparado con RAM, parecido a leer de un disco → latencia ≈ >10 ms → velocidad ≈ 50 MB/s La gran ventaja: S3 tiene millones de discos en paralelo → no es rápido por acceso individual → es rápido si haces muchas peticiones a la vez Si haces muchas requests simultáneas, puedes lograr muchísimo ancho de banda total Pedir objetos pequeños es caro → el coste y la latencia por request pesan mucho Objetos grandes son ideales para S3 → el coste por request se vuelve insignificante → compite bien (o mejor) que EC2 Otros servicios tipo S3 (Azure Blob, GCS, etc.) se comportan muy parecido

Cloud Information Systems Flashcards

(50 cards)