What is cloud computing?

What is cloud computing?

UT DALLAS Erik Jonsson School of Engineering & Computer Science Cloud Computing Trends FEARLESS engineering What is cloud computing? • • • • • ...

2MB Sizes 0 Downloads 27 Views

Recommend Documents

Cloud Computing
... Saugatuck Technology). ○. Web Services + Virtualisierung = Cloud Computing .... Versuch einer Definition). ○. â€

What is cloud telephony? - Polycom
Are you a small or medium size business? Are you part of a widely distributed, large enterprise? Are you tired of managi

Cloud Computing - Wikimedia Commons
Mar 10, 2015 - works: A revolution in the earth system science? ...... Grow to 26 Billion Units By 2020”. ... the Inte

Cloud Computing - Salesforce
Part of Prudential PLC, United Kingdom. ▫ First Foreign-Owned Consumer Finance Company in. Vietnam. ▫ Largest Financ

What You Need to Know About Cloud Computing - Gartner
of private cloud services), the migration will take decades, and even then quite a bit of IT will stay in-house; in fact

What is Duracell Cloud? - Sunstar Company, Inc
Duracell Cloud is the perfect solution for businesses that share files and synchronize folders across their network usin

Access Control in Cloud Computing
Discretionary Access Control (DAC). •. Role Based Access Control (RBAC). Now we have a lot of techniques for access co

Cloud Computing - Google App Engine
Cloud Computing - Google App Engine. Peter Sutter. Fakultät für Informatik. Hochschule Mannheim. Paul-Wittsack-Straße

Cloud Computing Timothy Grance - cendi
Posted by Bhavin Turakhia http://bhavin.directi.com/crowd-sourcing-harnessing-the-power-of-the-people/. Better Apps With

UT DALLAS

Erik Jonsson School of Engineering & Computer Science

Cloud Computing Trends

FEARLESS engineering

What is cloud computing? • •

• •

• • •

Cloud computing refers to the apps and services delivered over the internet. Software delivered over the internet is usually referred as Software as a Service (SaaS) – Salesforce.com – Google Calendar Cloud usually refers to data center infrastructure that runs those services Public cloud is shared by multiple organizations – Usually pay-as-you-go based access – Example: Amazon Web Services Private cloud is generally managed and owned by single organization Storage as a Service – Amazon S3 Platform as a Service (PaaS) – Microsoft Azure

FEARLESS engineering

What is new in cloud computing?

• Computing=utility? – Analogy to power utility • You do not built a power generator in your home • No need hire someone to take care of your in-house power generator • Pay as much as you use

• Three new aspects – The illusion of infinite computing resources – No up-front costs to use public clouds – Pay-as-you-go models

FEARLESS engineering

What is new in cloud computing?

FEARLESS engineering

Trends supporting cloud computing

• Mobile interactive applications – Respond to information provided by user and sensors in real time

• Rise of analytics and big data • Parallel batch processing – Hadoop, Map-reduce

• New business models – Pay-as-you-go

FEARLESS engineering

Emergence of Big Data





“Big data has now reached every sector in the global economy. Like other essential factors of production such as hard assets and human capital, much of modern economic activity simply couldn’t take place without it. “ Big data creates value – – – – –



MGI Big data report FEARLESS engineering



Creating transparency Enabling experimentation to discover needs, expose variability, and improve performance Segmenting populations to customize actions Replacing/supporting human decision making with automated algorithms Innovating new business models, products, and services

Big data will create different opportunities in different industries To scale to big data, cloud computing technologies will be critical

Cloud computing Infrastructure Variants: Computation Model

Taken from U.C. Berkley Technical Report FEARLESS engineering

Cloud computing Infrastructure Variants: Storage Model

Taken from U.C. Berkley Technical Report

FEARLESS engineering

Cloud computing Infrastructure Variants: Networking Model

FEARLESS engineering

Example: Microsoft Azure • Platform as a service • Provides tools for building SaaS • Software sits on Microsoft Data Centers

FEARLESS engineering

MS AZURE Architecture Overview

• • • • •

Compute: runs applications in the cloud. Those applications largely see a Windows Server environment, although the Windows Azure programming model isn’t exactly the same as the on-premises Windows Server model. Storage: stores binary and structured data in the cloud. Fabric Controller: deploys, manages, and monitors applications. Content Delivery Network (CDN): speeds up global access to binary data in Windows Azure storage by maintaining cached copies of that data around the world. Connect: allows creating IP-level connections between on-premises computers and Windows Azure applications.

FEARLESS engineering

MS AZURE: Compute

FEARLESS engineering

MS AZURE: Compute

• Developer needs to submit configuration to define various roles. • Each role will be assigned a VM by Fabric controller • Various internet protocols could be used for accessing roles • Load balancer could assign requests to roles arbitrarily – State information needs to be maintained through database

• Number of role instances could be dynamically increased or reduced. • Monitoring and debugging services are provided by Azure

FEARLESS engineering

MS Azure: Storage

FEARLESS engineering

MS Azure: Storage •









Blobs are used to store binary objects such as videos, images. – Similar to Blobs in databases – Containers can contain multiple blobs – http://.blob.core.windows.net// is a unique identifier assigned when a new storage account is created, while and are the names of a specific container and a blob within that container. – It could be public or private • For private, you need to authenticate by signing requests Tables – Similar to Google Bigtable, (many entities, each entity has properties, each property is (name,type, value) – Need to use Odata to query – Distributed storage to scale to big data Queues – Allows Web roles to communicate with Worker roles – Web role can issue a computation request – Worker role can return results using other queues Replication is done automatically – Three replicas similar to Hadoop – Back up copy of all data is kept in another DC Provides restful interface to access data – Simple Http calls are enough to access data

FEARLESS engineering

MS Azure: Fabric Controller

FEARLESS engineering

MS Azure: Fabric Controller

• • •





Controls all resources including load balancers, computer, switches etc. Each computer runs a fabric agent Handles load balancing and recovery based on the XML configuration provided by the developer – Moves role instances to different VMs for load balancing – Add new role instances if some of them dies – Allocates role instances to different size machines (e.g., extra small to, extra large) – Instances are grouped to prevent single point of failure Updates and patches for VMs running web and worker role instances are handled automatically – Assumes at least two instances are running for each role – Stops one for upgrades while the other instance is running Updates and patches for applications running on Azure, update domains are created. – Fabric controller stops machines in one update domain, updates the app and moves on to next domain.

FEARLESS engineering

MS Azure: Queues

• •

Web roles instance sends a message to worker role using queues. Work role instance deletes the message after it is done. – Why?



Not a typical queue. – Each message can appear multiple times – No first in/ first out semantics – No guarantees on ordering of messages

FEARLESS engineering

MS Azure: CDN

FEARLESS engineering

MS Azure: CDN

• Provides capabilities similar to traditional CDN • Caches blob content closer to actual users • Useful for delivering multimedia files • Enables better experience for end-users

FEARLESS engineering

MS Azure: Connect

FEARLESS engineering

MS Azure: Connect

• Allows connecting local resources with Azure • Need to run Azure endpoint agent on Windows • Connections is done through IP V6 • Once connected Azure application appears to run on the same IP network – Useful for connecting in-house databases with Azure roles

FEARLESS engineering

MS Azure: Scenarios • Allows a typical web app to scale up and down as needed – Increase/decrease the web roles as needed

• Easy to scale to large demands • All load balancing and management done by Fabric controller

FEARLESS engineering

MS Azure: Scenarios •

Parallel processing can be done easily by creating multiple worker roles – Note that running 10000 machines one hour, costs similar to running one machine 1000 hours



FEARLESS engineering

Queues are used to communicate messages

Economics of Cloud Computing

• Provides economies of scale in three areas: – Supply-side savings • Large scale data centers cost much less

– Demand-side aggregation • Smooths the variability

– Multi-tenancy efficiency • Maintenance cost is divided over multiple users

FEARLESS engineering

Supply-side Economies of Scale •

• •

FEARLESS engineering

Cost of power – Usually DCs are located in low power cost areas. – 1 KWH= 3.6 cents in Idaho versus 1KWH=18.0 in Hawaii Infrastructure labor costs – An admin can service thousand of computers easily Buying Power – Google pays less than you do for an Intel CPU 

Supply Side Economics

FEARLESS engineering

Demand Side Economies of Scale



Utilization is critical for efficiency – In non-virtualized world, typically each app runs on its dedicated server – Typically utilization is low



Low utilization reasons – Randomness (people check their Facebook pages at different times – Time of the day pattern (people watch Netflix in the evening more often) – Industry Specific Variability – Uncertain Growth Patterns

FEARLESS engineering

Demand side of economies of scale

FEARLESS engineering

Demand side economies of scale

FEARLESS engineering

Demand Side Economies of Scale



• • • •

Aggregating over multiple organization, applications, and industries may – reduce the variability in demand – Increase resource utilization – Prevent overprovisioning • For example, peak demand/average traffic approximately 4 in general retail Flexible use of money and resources – 1000 machines x one hour usage= 1 machine x 1000 hour usage Shifting the risk in terms of resource estimation to cloud No up front investment Pay-as-you-go models

FEARLESS engineering

Multi-tenancy Economies of Scale

• Fixed application labor amortized over a large number of customers – Cost is shared by multiple organizations

• Variability in demand could be further decreased by Multi-tenancy if the tenants are from different industries

FEARLESS engineering

Overall Impact

• Due to economies of scale, large DC may have up to %80 reduction in Total cost of ownership. • For some apps, moving to cloud can create significant cost savings.

FEARLESS engineering

Moving to Cloud?

• As we have seen in the Azure case, creating apps that can leverage cloud could be challenging • SaaS does not require such development – MS Office 365-S

FEARLESS engineering

Possibilities

• Elasticity could be important for some apps – Running an multiple instances at the same time – New massively parallel applications

• Elimination of capital expenditure – Critical for new start-ups – No need to buy infrastructure to create your next idea.

• Reduction of Complexity – You do not have to manage your infrastructure – You do not need to manage your on web server etc.

FEARLESS engineering

Obstacles

• There are many obstacles in cloud adoption • Possible perception issues related to cloud

FEARLESS engineering

Obstacle: Availability of Service • • •

Too big to fail in Cloud computing Need to run your apps on multiple providers to prevent single point of failure Opportunity: Cloud elasticity can make DDS attacks more costly – – –

– –

FEARLESS engineering

Suppose EC2 can handle 500 bots Attack generates 1GB/sec traffic using 500,000 bots costing 15000$ At AWS attack will cost 360$ per hour in bandwidth and $100 per hour in computation need 1000 EC2 instances) After 32 hours, cost of the defense is larger than attack Attacker needs to sustain attacks longer.

Obstacle: Data Lock-In

• Your data may die with the company – Linkup shut down after loosing 45% of customer data – It turns out Linkup used another service called Nirvanix.

• Once you lock into a cloud provider, they can increase the prices • Possible solutions – Standard APIs and tools for data and app migration – Support of Hybrid Models

FEARLESS engineering

Obstacle: Security, Privacy and Compliance



Possible security and privacy issues related to data that is pushed to public cloud – This will be our main focus during the rest of the class.

• • •

Will Amazon fight to protect your data against Government subpoenas? Compliance issues (e.g., HIPAA) Possible solutions – – – – –

Encryption ( we will spend four weeks on encryption related solutions New cloud auditing solutions VPNs in the cloud Application and VM firewalls Location aware data storage • Keep your data in European jurisdiction that provides higher privacy guarantees.

FEARLESS engineering

Obstacle: Data Transfer Bottlenecks

• Transferring large data (10TB) to Amazon may take long time on 20 Mbit/sec connection – 10*1012 / ((20*106)/8)=4,000,000 seconds approximately 45 day!!! – Overnight shipping of hard disk would be much faster 

• Possible solutions: – Keep all your data in the cloud. • Amazon now stores some publicly available data sets.

– Faster and cheaper WANs

FEARLESS engineering

Obstacle: Performance unpredictability

• If multiple VMs run on the same server, I/O performance can significantly vary. • Possible solutions – Better virtualization – Flash memory based storage (still somewhat expensive)

FEARLESS engineering

Other obstacles

• Storage systems that can easily scale up and down as the demand changes • Development issues (i.e., bugs) and movement costs • Scaling automatically and quickly – Scaling is currently manual and not trivial

• Bad guys using the cloud – Spams coming from EC2 – Hosting spam web pages on EC2

• Need for new software licensing

FEARLESS engineering

Private Cloud Architecture?

• Resources are pooled across company • Possibility of using inhouse versions of Azure

FEARLESS engineering

Cost Trade off public versus private • Private cloud is a cheaper option for large companies • Still analysis on the left does not consider the security risks and costs,

FEARLESS engineering

Other architectures?

• Here at UT Dallas, we advocate for hybrid cloud solutions – Sensitive data will be mostly kept in private cloud – Sensitive data in public cloud will be encrypted – Intelligent query processing techniques will be used efficiently combine the resources of public and private clouds.

FEARLESS engineering