Walter Dal Mut

github.com/wdalmut

twitter.com/walterdalmut

https://corley.it

What is cloud computing

Summary

• On demand - self service
• Anywhere, Any time, any device
• Location independent
• Elasticity
• Pay as you go

Cloud Computing Servicesvalue visibility

On Demand

Every Cloud provider exposes API
Application programming interface

It means that you can access programmatically
to any service exposed by the provider

API means expose services

Now the light bulb has an unique HTTP address

https://somewhere.someprovider.com/bulb/198357

Our service expose a way to work with that light bulb

• Turn it on POST /bulb/198357 {"state": "high"}
• Turn it off POST /bulb/198357 {"state": "low"}
• Get current state GET /bulb/198357 return {"state": "high|low"}

Anybody can develop applications!

• Turn it on/off at sunrise or sunset
• Emulate people presence in an house(useful when we leave on vacation)
• On alarms to turn it on/off
• There are so many possibilities...

So we can decouple our system to different and reusable parts

Reusable parts means cost saving

APIs immediately creates a new building block for any application

Now we can create our service oriented infrastructure

Pay as you go

Focus on scalability of you applications

Scalability

Scalability (Spike)

Scalability (daily hours)

Provisioning Hardware (o VM)

Real Usage

Money waste

Autoscaling environments

SOA - Service Oriented Architecture

Autoscaling - Price

Serverless

Pay for "Actual Demand"

Serverless

• No server to manage
• Scale on real usage instantaneously
• High Availability
• Pay only for single requests

Serverless + Api Gateway

Serverless beyond computing

Pricing example

5 milioni * 3,50 USD/milione     = 17,50 USD

3 KB * 5 milioni = 15 milioni KB = 14,3 GB

14,3 GB * 0,09 USD               = 1,29 USD

17,50 USD + 1,29 USD             = 18,79 USD

                    

0 requests => ~0 USD

Scalability Summary

• Scale capacity on demand
• Turn fixed costs into variable costs
• Always available
• Rock-solid reliability
• Cost-Effective
• Reduce time to market

Monolith

Problems

• Difficult to scale
• Long development cycles (dev, build, test)
• So many modules (operations nightmare - who is the owner?)
• Architecture legacy (hard to maintain and evolve)
• Add features are difficult for developers
• Feature delivery can be really slow

It means...

• Lack of agility
• Lack of innovation
• Frustrated customers

Services to Micro Services

Composing and collecting APIs create a new business model where an application is decoupled into many tiny little services

What is a microservice

service-oriented architecture composed of loosely coupled elements that have bounded contexts

Adrian Cockcroft @ Netflix

Create many tiny services

Instead doing one thing

Do one thing, and do it well

Microservices achievements

• We can scale the development team
• Easy deployments (much more deployments)
• Reduce infrastructure costs

Delivery/Deployment...

hundreds of teams      x
thousands of services  x
different environments x
deployment time        = ?

If we take 1 hour to manually build/test/deploy

Continuous delivery/deployment

Delivery automation

If we have 300000 deploy per year
our delivery pipeline must deploy every
30 seconds

AWS CodeBuild/CodePipeline

Measure is the key to science

Now we have tiny little services and
we can deploy those services easily

How we can save resources and deploy more applications in consolidated environment?

Instance resources

More applications?

resource contention

In computer science, resource contention is a conflict over access to a shared resource such as random access memory, disk storage, cache memory, internal buses or external network devices.

Containers

Lightweight, isolated environment for applications

CAP - Limits for resources (CPU, Memory, Networking)

It is like a sort of virtual machine but it is not!

if with virtual machine we can run up to 15/20 instances per phisical hardware, now with containers we can spawn hundres if not thousands of application per phisical hardware

Distributed architectures

Clusters for web applications

Microservices

More instances, more resources

More instances, more resources

Api gateway for applications

We have this sort of super computer where we can run our applications

Kubernetes & Docker Swarm

Orchestrators for Containers

Export control with Cloud & On-Prem federation

Many isolated but somehow connected clusters for appplications

Big Data

Big data is a collection of data sets

• so large and complex
• difficult to process

Big Data Challenges

• Capture
• Storage
• Search
• Share
• Transfer
• Analysis
• Visualization

Common Scenario

• Data arriving at fast rate
• Data are typically unstructured or partially structured
• Data are stored without any kind of aggregation

Data Grow

• Scalable storage
  • In 2010 Facebook claimed that they had 21 PB of storage
    • 30 PB in 2011
    • 100 PB in 2012
    • 500 PB in 2013

Massive Parallel Processing

In 2012 anybody can elaborate exabytes (10^18 bytes) of data in a reasonable amount of time

Reasonable Cost (Cloud Computing)

The New York Times used 100 Amazon EC2 instances and an Hadoop application to process 4TB of raw images into 11 million finished PDFs in the space of 24 hours at a computation costs of about $240

Big Data examples

• So many data (petabyte of information)
• So many different not aggregated data sources
• ...

Petabyte of information

We cannot create a terabyte large index of our data (doesn't fit our memory)

It means that we hit a phisical limit for traditional tecnologies (Relational Databases)

Different/Not Aggregated data sources

We cannot join not structured data and from different data sources

How we can join information in our service oriented infrastructure?

We can solve those computing problems introducting a way to analyse datasets

Map/Reduce

Map

A mapping is a transformation from one value to another value

[1,2,3].map(multiplyBy3) => [3,6,9]
                    

For example, if you start with the number 2 and you multiply it by 3, you have mapped it to 6.

Reduce

This operator combines mapped values down to the final resultset

[1,2,3]
    .map(multiplyBy3) // [3,6,9]
    .reduce(sum)      => 18
                    

Extend to parallel computing

Map/Reduce is an highly scalable data processing algorithm

it is all about brute force

Hadoop is a framework for distributed map/reduce

Thanks to Hadoop ecosystem we can use higher level tools to analyze/visualize etc... our data

Apache Hive

Run SQL statements over our distributed data source

Well-Known SQL statements

SELECT * FROM tickets t WHERE t.weekday = 12;
SELECT * FROM tickets t GROUP BY weekday;
SELECT * FROM tickets t JOIN users u ON (t.owner_id = u.id) WHERE u.id = 12
etc...
                    

Blockchain

Blockchain and not Bitcoins

Bitcoins are just an application of a blockchain but it can help us to focus on different things about blockchains

How maintains the ledger?

Central authorities in general: banks, etc.

Why we want to drop central authorities?

Transparency on transaction/tracking

...

So drop central authorities

Drop the central authorities

Do we spot how many problems do we have now?

• Privacy problems
• Fraud transactions
• stale/not updated ledger
• ...

Privacy problem

Otherwise everybody knows my balance

How we can solve other problems if everybody has complete access to the ledger?

The way that blockchain uses to provide security is
TRUST NOBODY

Blockchain provide mathematical and crypto challenges to achieve that results and protect itself

How to maintain the DISTRIBUTED ledger?

Blockchain do not store a computed ledger but transactions

Transaction wrap an information

The ledger is just a concatenation of transaction

why is blockchain and not transaction chain?

because transactions are groupped into blocks
(of transactions of course)

Blocks are linked together to create the chain

Here the blockchain

How we could prevent data modification of a block?

How we could prevent that anybody send fraud trasactions?

With asymmetric cryptography

Digital Signature

Public Key + Private Key

• Only the private key can sign a transaction/block
• Everybody can verify a signature of a transaction/block with the public key

Digital Signature

How to link blocks together?

Every block has also a previous hash

Who introduce new blocks in a distributed network?

Everybody try to create blocks

The first one that creates a new block in the network is the winner

To win a reward and create a new block for the chain everyone in the network has to solve a complex mathematical problem

Proof of work

Solve a mathematical problem

Hash function of something start with `n` zeros

sha256(1404) = 000ebec3ecebd727ce4f020441268ad34ac468cdb1c76ced442380b5ac842b7d

Complex enough to take 10 minutes to solve the problem (bitcoins)

So everybody try to "mine" a new block...

What happend if two actors compute a valid block at the same time?

This is a real problem because we can have a big issue!

Double Spending Problem

Longest chain always win

Double spending on the latest block
wait 6 blocks ahead yours to have a strong confirmation of your transactions

Orphans and blockchain

The chain must be continuously incremented to provide stability

Empty blocks or transaction filled blocks are equivalent, the point is minimize the double spending problem

But if we have to wait 6 blocks ahead our transaction to be reasonably sure about confirmation we have to wait 60 minutes (10 minutes per block generation)

Tipically we wait 2 business days to move money between bank accounts!

Private Block chain

In a private blockchain we want all features but we are not interested in rewards and proof of work in general, instead we want more speed for inserting blocks

Proof of Stake

Proof of stake is a different way to validate transactions based and achieve the distributed consensus.

The creator of a new block is chosen in a deterministic way, depending on its wealth, also defined as stake

walter.dalmut @ corley.it