L6: Applications and Paradigms

Cloud Application:

focus mainly on enterprise computing
Ideally, an application can partition its workload into n segments and spawn n instances, and the execution time reduced by a factor of n
Key challenges:
- Cloud consumer: scale application to accommodate a dynamic load, recover after a system failure, efficiently support checkpoint/restart
- Cloud provider: manage a large number of systems (cloud consumer applications), provides quality of service guarantee
Ideal applications: workload that requires massive computing infrastructure
- web services, database services, search services (web, images, etc), ML with massive-scale models
Unlikely to perform well:
- Applications with a complex workflow and multiple dependencies such as high-performance computing
- Applications with intensive communication among concurrent instances
- Workload cannot be arbitrarily partitioned

Performance isolation: how to ensure that customer performance is not affected by other users
Reliability: major concern, server failures expected when a large number of servers cooperate to compute

→ Checkpoint/restart, replicate service, autoscale and load balance
Cloud infrastructure exhibits latency and bandwidth fluctuations which affect the application performance
Performance considerations limit the amount of data logging (identify unexpected results, errors, monitor application performance)

Reliance on Internet and Web technology (high accessibility, web browser universality, ease of web-based service development)
Cloud services use web technology as both the implementation medium and management interface
2 basic components of the web are web browser client and web server, i.e. based on client-server architecture

Architecture of web applications: 3-tier architecture

Presentation layer represents user-interface → sits on both the client and server side
Application layer represents implementation logic → web server receives client requests and retrieves the requested resources or generates web content based on application logic
Data layer comprises of persistent data stores → web server interacts with application servers and the underlying databases

→ PaaS typically provide separate instances of the web server, application server, and data storage server environments

Web Services technology: provides “as-a-service” cloud delivery models

Simple Object Access Protocol (SOAP): application protocol for web applications; defines a common web service messaging format for request and response message exchanges; based on the XML, uses TCP or UDP transport protocols
Representational State Transfer (REST): software architecture for distributed hypermedia systems. Supports client communcation with stateless servers, platform and language independent, supports data caching, and can be used in the presence of firewalls

Cloud services are naturally distributed → a single service request might require message passing between multiple servers
Cloud services are accessed by different clients using different programming languages and technologies → requires a standard interface for communication between clients and server
Web Service APIs implement a standard communication interface for cloud services, e.g. REST API requests/responses are transferred on HTTP in formats such as SML and JSON

Untitled

Characteristics of IaaS:

Consumer:

Provider:

Characteristics of PaaS:

Application is located on the cloud
Clients use a thin client (web browser) to access the program
Often built on top of multi-tiered PaaS and IaaS
- Clients do not deal with underlying PaaS and IaaS
- Clients do not deal with APIs
Various pricing mechanisms: free, pay-as-you-go, freemium, etc

Characteristics of a SaaS:

What? Build scalable event-driven applications

Why? Relieves cloud consumer server management (proisioning, scaling, and monitoring, etc), no idle capacity

How?

Developer focuses on application/business logic, application is divided into smaller, single purpose (Lambda) functions
Event triggers a Lambda function and dies after execution
Lambda is stateless
No VMs/containers in the programming models
Vendor provides provision-free scalability
Use cases: web/mobile backends, real-time data processing, chatbots, virtual assistants, etc

Lambda function contain your application logic (your code)
Configuration defines how your code is executed, i.e. configure event sources that invoke your functions
E.g: API Gateway event source can invoke a lambda function on incoming HTTPS requests
Taken care by system: code to integrate event source with lambda function, manage infrastructure that detects events and invoke lambda, etc

Supports arbitrarily divisible workload
Supports distributed computing on large datasets on multiple machines (clusters, public or private clouds, etc)
How large an amount of work?
- Web-scale data on the order of 100s of GBs to TBs ot PBs
- Input data set will not likely fit on a single computer’s hard drive
- A distributed file system (e.g. Google File System GFS) is typically required

Untitled

Key idea: (a) split data into blocks and assign each block to an instance/process for parallel execution (b) merge partial results produced by individual instances after all instances completed execution
Transform a set of input <key, value> pairs into a set of output <key, value> pairs
SPMD → Same Program Multiple Data