Lecture 9

• Lecture 9
  • Slides
  • Video
  • Serverless Computing
  • Programming Model
  • Platform as a service vs Serverless Computing
  • Serverless Comp. vs Microservices
  • Service Level Agreements (SLAs)
  • Reference
  • Distributed Computing - Clusters
  • Objectives of Cluster
  • Cluster Components
  • Cluster Classifications

Slides

link1 link2

Video

link

Serverless Computing

• does not mean servers nahi hongey
• it is just that developer ko yeh sab nahi sochna, woh cloud provider dekh lega
• u just write functions that handle incoming events say.

• AWS me lambda is serverless computing
• Container me functions execute hote, but that scheduling cloud wala dekhta
• these functons are stateless

• there are execution logs

Programming Model

• for every event a new instance is executed, so if u want to store smth, write it to permanent storage

• boto3 is AWS’s SDK
• lambda lightweight service, much used
• serverless computing is closely related to microservices.
• programmer does not worry about how to allocate resources, how many functions to run etc.
• the functions are stateless so can run in parallel, they do not share state
• they might share db state, but that db will take care of

Platform as a service vs Serverless Computing

• Paas me bhi api de deta provider, and abstracts out things
• diff is in terms of stateless functions
  • in paas, whole programming env is stored
  • every event that comes can see cumulative state of environment
  • in serverless compu, state is stored in db and not runtime env
  • and, we can write any code w/o cloud provider packages
  • so I can shhift it to any other cloud provider
  • i.e. loose coupling here

Serverless Comp. vs Microservices

• look up aws lambda, google cloud functions, microsoft azure ka bhi

Service Level Agreements (SLAs)

• any service a provider offers, a agreement goes on b/w cloud and service provider
• SLAs have aspexts about service availability, performance, security, privacy etc

• there are quantifiable metrics as well called Service Level Objectives
• and if cloud provider does not provides it at any point of time, paisa bharegea

• availability rate = % time it was up

• reliability rate is only applicable to Software as a service model

• instance starting time = new VM ko start hone me max kitna time, avg kitna etc
• response time is for synchronous services
• completion time is wrt asynchronous service

• resilence = recovering

• switchover is kind of backup thing, primary fail hua, switching to second
• recovery is like system is down, kitni der me wapis a jayega

Reference

Distributed Computing - Clusters

• a clustyer is collection of stand alone computers on which we do some parallel processing

• pipelining
  • instruction level
• super scalar
  • having multiple pipelines, divide instrns such that no overlap b/w piipelines
• multi threaded
  • single pipeline exposed as 2 diff processors to os, and os schedules
  • ek stall hua toh dusre se lele
  • pseudo parallel - not visible to programmer
• multi-core, many-core
  • visible to os
  • it can schedule accordingly
• shared memory over custom interconnect
  • shared memory hai
• message pasing over custom interconnect
• clusters
  • systems connected over a n/w
  • not generally high-end systems
  • normal pcs le lie markets se
  • easy to use distributed parallel computing
• grids
  • sharing resources across multiple administrative boundaries
  • another laboratory ko connect kr raha and using resources
• cloud
  • it is pay per use thing
• parallel is within the system
  • I can process job within system
• clusters
  • intersection of parallel and distributed
  • parallel in terms of algorithms
  • distributed => multi-tasking system
    • bahut kam hai, har system ko kuch kam de de
    • Hadoop ki tarah
      • the ndoes ar not executing some algo

Objectives of Cluster

• high performance
  • running on multiple systems
• high availability
  • redundancy
  • storing same data or running same service on multiple systems

Cluster Components

• high speed n/w = infiniband, gigabit ethernet or smth
  • to enable faster communications b/w nodes
• nodes also have os on top of that
• on top of os, we have runtime environment
  • where eveloper writes code and executes
  • cluster middleware implements whatever parallel programmig task is to be done

Cluster Classifications

• dedicated => all nodes within cluster are avilable for cluster only
• sometimes ek system pe kuch aur bhi kam ho rah eho sakte
• os in gen sabme same hota
• homogeneous => same congiguration and os sabme
• level of clustering
  • group
    • custom switch = custom build h/w