Semester7

Notes of courses done/attended in semester 7 in college

before cloud computing existed, Software as a service was still existing
what makes cloud computing -
- provides illusion of infinite resources
  - as and when necessary, I will take more servers/resources
  - I do not have to predict max demand already
- Commitment required by cloud users (who use the services)
- ability to pay for use of resources on short term basis as needed
  - abhi liye, and thodi der me returned
eg
pehle wale failed bcz I need to negotiate and sign contract and all
but successful wale
- billing is per hour basis or less
- scalable storage service (S3)
- no contract and all
- bandwidth charges are very less as well
How these services are offered so cheaply
when u do at big scale, u can do at cheap rate - economies of scale
u can make multiple instances into single box
- put multiple instances as virtual boxes/containers into single box - statistical multiplexing
bulkpurchasing
- discount jyada when u order things in bulk - eg
is it possible for small scale company to provide cloud services
- Amazon toh kar sakta
  - they have s/w stack to manage huge setup
  - and infrastructure as well
  - they extended it to public
  - utilized off-peak capacity
    - rented things they were not using
- Microsoft
  - had lot of consumer base who used office suite and .NET products
  - it just moved them to cloud
- Google
  - had large setup and s/w stack
  - used that to offer cloud services

user

u cannot give up capacity in static data center tho demand is varying
- increases cost
Cloud side - resources are changing according to demand

eco2

when I have undercapacity, I will be dropping requests which exceed the capacity
- so I will lose confidence of users, demand will decrease (see 2nd graph)
so ppl prefer cloud computing as a platform

cloud

prov

multi-tenancy
- m consumers will be using ur system simult maybe
- m se multiple nahi kia but function of m
- function will be sublinear
I am multiplexing
I will have some metadata corresponding to each customer
summing up costs , it should be less than total paisa consumers pay, tbhi profit hoga na bhai
Analogies
- hotels
  - multiple ppl check-in/check-out
  - one room shared by multiple ppl
- time-share/vacation homes

attr3

resource pooling
- multiple consumers can access the resources
- provider’s resources are pooled
- location of resource is not reqd to be known by consumer

attr2

diag

deployment models
- community cloud
  - gp of companies together using a cloud
- hybrid
  - combn of public and private
Delivery models
- SaaS - software
- PaaS - platform
- IaaS - infrastructure
- DaaS - database

Cloud applications
Parallel Programming paradigms
- pleasingly synchronous model
- bulk synchronous model
  - hadoop

stream

streams = continuous generation of data
items come 1 by one, and are timestamped
2 challenges
- very fast
  - processing of one item must finish before other arrives
    - then only we can do without loosing buffer and all
- and very large
  - it’s unlimited stream
we need to extract info realtime
- cannot store everything at one place and then apply algos
also, data might be evolving
- so models which I build, should adapt to change in data

ddf

ActiveMQ
- a queuing system
- generator can put stream into queue and reach destn
- scaling up is prob
Scribe
- open-sourced by FB
- pools data and stores into disk
Flume and Kafka
- recent open source systems
- meant for distributed collection of data
- follow publish/subscribe model

ddf2