transformation to the cloud...image migration –capture an image of the server and reconstitute in...

Nikolai Joukov

John J. Rofrano

Murthy V. Devarakonda

Harigovind Ramasamy

Transformation to Cloud Computing

New York University, Course CSCI-GA.3033-011, Spring 2015

Outline

Background on the course

The approach to transformation

Course Details

2 Transformation to Cloud Computing, New York University, Spring 2015

This Course is a Bit of a Horse of a Different Color

This course it being taught by a team of IBM Researchers who are experts in their field

There is no text book because “the book” is still being written ;-)


Transformation to Cloud Computing

4

What is Cloud Computing?

…and why would anyone want to transform to it?

Examples of Cloud Computing

5

Utility Computing

Computing as a Service

Shared Resources

SaaS,PaaS,IaaS

On Demand Provisioning

Cloud is… All of The Above


Some Cloud Infrastructure Services


Some Cloud Platform Services


Some Cloud Software Services


What we mean when we say “Cloud Computing”

A fundamental shift from physical infrastructure to virtual infrastructure– Seen as a major paradigm shift

A rapid way of provisioning and later releasing computing services on the network

Some important characteristics:– Rapid/automated provisioning and (later) release of services– Can be Pay-as-you-go– Appearance of infinite resources– Could be managed or unmanaged– and more

Types of Clouds:– Based on service:

• Software as a Service (SaaS)• Platform as a Service (PaaS)• Infrastructure as a Service (IaaS)

– Based on ownership• Public Clouds• Private Clouds• Hybrid Clouds


Transformation

11

What is Transformation?

Transformation is Migration

Since the cloud is virtual you cannot “move” a “physical” computer into a cloud

– But you can move the disk image into the cloud (which can be just as good)

• …unless you had lots of “junk” on your computer because you might not want that on

your new cloud server (note: we call this “your mess for less”)

– Or you can build a new server in the cloud that looks just like the old server

• When we build a new server we sometimes call this “workload” migration because

we are migrating the “work” that the server is doing into another server in the cloud.

What if the cloud doesn’t support your server’s OS platform?

– Now you have to “re-platform” to a new OS which could get tricky

– This is a lot like building a new server but what about all of the software? Will it still

work?


Types of Migration

Image Migration– Capture an image of the server and reconstitute

in the cloud• P2V (Physical-to-Virtual)

• P2I (Physical-to-Image)

Application / Workload Migration– Build a new server to run the workload in the

cloud• Re-install

• Re-platform

• Consolidation

– Build from a Template• IAAS

• PAAS

– Build via Specification Cloning• Use the physical server specs to build an exact

duplicate virtual server

ImageMigration

WorkloadMigration


Types of Clouds

Unmanaged Clouds– Do-It-Yourself from scratch

– Possibly bring your own image

– Anything goes but you fix it when it breaks

Managed Cloud– Assistance in on-boarding

– Only certain image infrastructure is allowed

– Bring your own image may not be supported• Or image compliance must be adhered to

– Steady state is managed• Patch management

• Upgrade management

ImageMigration

WorkloadMigration

ManagedCloud

UnmanagedCloud

14

More Types of Clouds

Clouds can be Public, Private, or Hybrid– Public to all on cloud provider’s infrastructure

– Private to business on dedicated business

infrastructure

– Private to business on shared cloud provider’s

infrastructure

ImageMigration

WorkloadMigration

ManagedCloud

UnmanagedCloud

15

HYBRID

PUBLIC PRIVATE

A transformation example you may have seen

16 * Example from Apple http://www.apple.com/support/switch101/

http://www.apple.com/support/switch101/

A transformation example you may have seen

17

Know your Target

Migrate what you can

Test and fix Problems

* Example from Apple http://www.apple.com/support/switch101/

http://www.apple.com/support/switch101/

Why switch PC platforms?

Will it save me money?

Will I have fewer crashes?

Will all my software work?

Will the things work that I need?

Will the service be better?

18

Why switch to cloud? A Business Perspective

My machines, my OS

My management tools, my

management processes

My delivery model

19

Standardized OS and no worries about machines

Standard set of tools, standardized processes

Standard Cloud Delivery model

My mess for less All good and cheaper in the Cloud

If costs will be significantly lower in Cloud offerings compared to current costs, will businesses tolerate running two models in parallel?

Initially this may be the case

The future of business computing

environments is likely to trend towards

sourcing from Cloud providers.

How will IT service providers manage the

desire from businesses to have cost benefits

of standard delivery models while retaining

some level of customization (some of which

is driven by industry/geo specific regulation)

The transition is challenging and businesses

want to understand how to do this with a

short time to payback

20

21

Corporate legacy has positive and negative sides and forms a legacy value stack

Legacy can be heritage or burden.

Technologies can help customers to identify

which legacy is heritage and which is burden,

and help to take the appropriate actions to

handle legacy.

Corporate legacy cuts across every

aspect of the enterprise.

Therefore, it can only be addressed effectively if

it integrates Systems, People & Culture, and

Processes around an agile business and

information design

Want to reduce,

replace, control

& get rid of it

Legacy

Heritage

Want to

leverage &

expand on it

Burden

“Good”

Legacy

“Bad”

Legacy

Incre

asin

g v

alu

e

Business model legacy

Business process legacy

Software & data legacy

Infrastructure legacy K

no

w-h

ow

le

ga

cy

Future of Legacy

Legacy needs to be treated differently depending on the heritage/burden aspect.

Any legacy related effort has higher value when targeting higher items in the legacy stack.

Transformation is Not Easy

Production workloads are running and

can’t be disrupted

System structure never fully known

– Personnel that knew the systems

changes

– Documentation gets out of date

Systems very heterogeneous

– Several generations of technology

– Line-of-business governance for

distributed systems

– Mergers and acquisitions

Hence transformation cost often exceeds the

expected benefits, and important

transformations are not undertaken. Big

area for improvement!

22

Outline



Course Details

23

Conceptual Migration Workflow

24

Know your Target

Discover your

Source

Analysis

&

Design

Provision &

Configure

Migrate & Remediate

Test Cut-over

Transformation Workflow Overview

25

Collect the c lient

technical, application, and financ ial data

Discove r Map Prov ision M igra te

V MMV M1V M2

D BA pp

S ourceV MMV M1V M2

D BA pp

Target

user

V MMV M1V M2

D BA pp

S ourceV MMV M1V M2

D BA pp

Target

user

Test

Leverage User transactions to

automatically test and compare target servers to source

Application and script transformation to

target O S

Create micro des ign for

selected sources and targets , us ing discovered data as reference

Create provis ioning request for net

new or migrated servers

Build, provis ion, and configure the target

platform( s ) environment and application components:

•O S •Middleware•Filesystems

•Users•H ardware•N etwork

Perform migration of

source middleware configurationto target

environment( s )

Ana lyze

Migration

Unit

Application

Data

Presentation

Migration

Unit

Application

Data

Presentation

Migration

Unit

Application

Data

Presentation

Migration

Unit

Application

Data

Presentation

Migration

Unit

Application

Data

Presentation

Migration

Unit

Application

Data

Presentation

Select migration

candidates

Create macro des ign( s) ,

RO I , and plan with c lient

Tool assisted framework for cross platform migrations

What is this Course About?

Enterprise IT needs constant changes:

– New business needs

– Reduce cost by new technology

Examples:

– Servers consume about 1.2% of the electrical power produced in the US

– Many servers have around 10% utilization – very inefficient

– Technology often gets 10-15 years old because upgrade too risky

Gradual transformation vs complete overhaul

We present approaches for typical enterprise IT transformation problems

26

Utilization of 9 servers in 24 hours (30-min slots), scale on the right

From An ROI perspective – successful enterprise IT transformation reducing risk/cost of migration and increasing steady state benefit

27

Steady

state

cost

benefit

Reduced

Power

Consumption

Reduced

Power

Consumption

Improved

Operational

Processes

Improved

Operational

Processes

Co

st

-

Improved

HW/SW

Utilization

Improved

HW/SW

Utilization

Source System Costfor steady state

Target System Cost for steady state

Design&

Approach

Planning&

Implementation

Realization

Cost

Migration

28

Abstract Architecture for Enterprise IT Transformation

Customer and

infrastructure data

collection

Consolidated data repositories

Analytics, Design,

Planning

Procurement and

physical setup

Non-customer

sources

Migrate and

modify

Test and

remediate

Synchronize and

cutover

Project workflow and status visualization

Migration

Analytics

29

Detailed Architecture for Enterprise IT TransformationCustomer and infrastructure data collection (manual and automated)

• Infrastructuredata collection

• Organizationaldata (owners ...)

• Customer goalsand plans

Consolidated data repositories

•Customer data

•Historic data

•General technical and financial data

•. Models

Analytics, Design, Planning

• Target options

• Technicalcompatibility

• Businessconstraints

• Placementoptimization

• Wave planning

• ROI analysis

Procurement and physical setup

• Hardware andsoftware ordering

• Setup of physicalinfrastructureif new

Non-customer sources

• External productdescriptions

• Benchmarks

Migrate, modify

• Core migrationmethods:

• Lift-and-shift

• Copy

• Provision and backup-restore

• Apply changesfrom design

• Start

Test and remediate

• Unit tests

• Comparison tests

• User acceptancetests

If fails, back to “modify” or initial state

Synchronize and cutover

• Educate

• Resynchronizetest system withsource

• Switch operationover

• Registrations etc.

Project workflow and status visualization

Benchmark assessment

Macro design

Micro design

Possibly retry

Possibly several optimization phases

For large migrations in waves = groupings migrated at different times.

Possibly even micro-design and procurement in these wavesContracts

Migration

Analytics

• Operational

Improvement

Case Study for this Course

30

Case Study Details

ACME wants to migrate some of their IT Services to the cloud

They “think” they have about 2000 – 3000 Servers running a

variety of:

– Linux 64-bit, 32-bit

– AIX

– Windows Server 2000, 2003, 2008, 2012

– Sun Solaris

Running middleware such as:

– DB2

– Oracle

– MySQL

– WebSphere Application Server

– JBOSS

– Tomcat

– SAP

In support of applications like:

– Company Web Site

– Human Resources

– Manufacturing

– Shipping and Logistics

– Custom Written Departmental Applications31

Outline



Course Details

32

Course Evaluation

50% Project–20% Project Presentation

• 10% Mid-semester presentation

• 10% final presentation

–30% Project Deliverables

20% Mid-term Exam

20% Final Exam

10% Class Attendance

34

Schedule from Start to Mid-term Exam

Lecture Title

Lecturers Date Description

1. Transformation To

Cloud Computing:

Course and Projects

Overview

Nikolai Joukov, Murthy

Devarakonda, Hari

Ramasamy, John

Rofrano

02/02 Introduction to cloud, IT transformation, and need

for transformation in enterprise infrastructures.

Description of course projects.

2. Enterprise IT Today,

IT Discovery

Nikolai Joukov 02/09 Enterprise IT today is a mix of technologies from

80s, 90s, 00’s, and state-of-the-art technologies.

Discovery is the first phase in almost every

transformation: Finding out the as-is state.

3. Clouds and Other

Transformation Goals

Murthy Devarakonda 02/23 Economics of cloud, types of cloud

4. Workload Placement:

Analytics, Design, and

Planning

John Rofrano 03/02 Analysis to identify which workloads are candidates

for migration to cloud. Algorithms to optimize

placement of such workloads on the target cloud

platform.

5. Performance

Baselining, Analysis,

and Verification

Hari Ramasamy 03/09 We’ll cover tools and techniques for identifying and

meeting performance objectives during migration.

6. Automatic migration John Rofrano 03/16 Automation technologies for physical-to-virtual

migration, re-installation, and re-platforming of OS

instances and applications.

7. Mid-term 03/30 Grades deadline is 04/06

35

Schedule from Mid-term Exam to Final Exam

8. Mid-semester student project presentations 04/06 Students will have the opportunity to report on their

project progress and receive early feedback.

9. Big Data

approaches to IT

Analysis

Murthy

Devarakonda

04/13 IBM Jeopardy!-winning solution and its uses for

enterprise IT

10. Network

optimization

Hari Ramasamy 04/20 Most transformation activities will involve

changes to the networking infrastructure.

We’ll cover automation techniques for

network reconfiguration during

transformation to cloud.

11. Enterprise Storage

Systems, DevOps

technologies to

maintain IT consistency

Nikolai Joukov 04/27 Deep dive into Enterprise Data Storage

Systems.

12. Testing Nikolai Joukov,

Murthy Devarakonda,

Hari Ramasamy,

John Rofrano

05/04 Testing is an important and costly phase in

transformation project. Testing techniques

used in transformation are quite different

from those used in software development.

13. Student Project Presentations 05/11

14. Final Exam 05/15

Project

Each student is expected to work on a semester-long project in a team of size 2-4

Project can be in any of the lecture topics

Choose a project topic, contact lecturers for that topic -- they will become your mentors

Mentors will help you define the problem, guide you as you seek the solution

Some projects involve implementation and/or experimentation, others don’t

You will be graded based on overall success of your project (teamwork is important) as well

your individual contribution

Identify 2-3 project topics that interest you, find a team, contact potential mentors

“Sign up” for the team and project by the 4th Lecture

Update your mentor periodically about your project progress, don’t wait until end of semester

Suggested Project Topics

Project: Enterprise Systems Compatibility Diagrams

Problem:– Diverse set of servers, storage, networking systems, and software– Discovery and management software has to be executed on these diverse servers– What server/OS, storage, and networking system pairs are compatible and to what extent?

You will learn:– Enterprise servers/storage/networking systems and middleware specifics

Expected results:– Diagram of systems lineage (at least 1995-2015)

• One person is expected to pick one of:• 1. servers + handheld devices• 2. storage systems• 3. networking systems• 4. databases, application servers, messaging systems

– Mark lineage/families/milestones (e.g., 64-bit platform)– Mark systems with OS types that they can run (e.g., Windows-based OS on some storage systems)– Mark software with hardware families they can run on (e.g., System Z)– Specify hardware details

Grading:– Expect to have no major vendor, architecture, or milestone missing for high grades – diversity is key

• Even if some companies do not exist today they should be included– Depict lineages and vendors mergers (e.g., what is the lineage of IBM XiV?)– Diagrams should be printable on just a few pages and presented clearly

• Discovery results presentation and clarity is a big factor so be prepared to spend time on it• You will have only several minutes to present your work so the diagrams should be self-explanatory

– You will be compared to all other students so make sure that the amount of work and details is appropriate

• Specify the sources of information

Mentor: Nikolai Joukov

Project: Transformation Decisions using Social Media Analytics

Problem:

1. There are many ways to buy a product or a service

2. Social media has a major impact on how people buy today, and so why not apply social media analytics to cloud transformation

decisions?

3. The task here is to develop a small prototype project to show how social media can be used for determining the right

applications and/or cloud providers for transformation

You will learn:

How to collect and prepare data for such an experiment, how to (modify if necessary but essentially) use an

existing sentiment analysis techniques to analyze the data and discuss/present results

Expected steps:1. Identify and collect relevant social media data2. Prepare the data.3. Identify and install sentiment analysis code and learn how to use it4. Run the analysis code on the data5. Organize, discuss, and present results6. Improve the code for better results if there is an opportunity to do so

Expected Results– Make it all work– Partial credit for completing part of the steps (say 1 – 4)

Grading:– Most of the grade (~85%) is for steps 1 – 5– Creativity and resourcefulness to make it all come together is the key to getting a high grade – think

outside the box!

Mentor Murthy Devarakonda, [email protected]

Project: Workload Placement Analytics

Problem:

1. Current workload placement papers focus on CPU utilization, while admitting that it’s a multi-dimension problem with memory,

IO, and possibly disk. Extend the algorithms to that.

2. In theory, the more dimensions, the less precise the heuristics. In practice, one tries to use time series (e.g., match morning-

intensive workloads with afternoon-intensive workloads), which adds dimensions. Try out whether it really gets better.

3. Our “main paper” (Mehta et al., next slide) considers servers only, the “more complex case” (Korupolu et al.) servers and

storage devices. How do the second algorithms behave if storage is no problem? (I.e., could we always use them? Could we

add ideas from the first algorithms so that they become optimal in that case too.)

You will learn:

Workload placement heuristics variants and gain experience how the heuristics behave for them.

Expected results:– Implementation of a multi-dimensional workload placement heuristics in programming language of your

choice.– Experimental design to demonstrate impact of varying one dimension over others.– Graphical display of the impact above.– Report on architecture and design of the approach

Grading:– Demonstration of the heuristics analysis– Regular progress meetings with the mentor– Clear definition of contributions of individual team members.– You will be compared to all other students so make sure that the amount of work and details is

appropriate• Specify the sources of information

Mentor John Rofrano

Closest lecture: Week 5, workload placementReading list:

– Basic terminology: First 1.5 pages each of:• L.T. Kou, G. Markowski: Multidimensional bin packing algorithms; IBM J. Res. Develop, Sept.1977, 443-448• Richard Karp, Michael Luby, A. Marchetti-Spaccamela: A probabilistic analysis ofmultidimensional bin packing algorithms; STOC 1984, 289-298

– Main paper: Sameep Mehta, Anindya Neogi: ReCon: A tool to recommend dynamic serverconsolidation in multi-cluster data centers; NOMS 2008, 363-370

– A more complex case: Madhukar R. Korupolu, Aameek Singh, Bhuvan Bamba: Coupled placementin modern data centers. IPDPS 2009: 1-12

– Related papers corresponding to other lectures, for discussion only:• How do we get the data needed here? N. Joukov, B. Pfitzmann, H. V. Ramasamy, and M. V.Devarakonda, Application-Storage Discovery, The 3rd Annual Haifa Experimental SystemsConference (Systor'10), Haifa, Israel, May 2010• A similar problem for different migrations: Andrzej Kochut, "Power and Performance Modeling ofVirtualized Desktop Systems", IEEE MASCOTS 2009• A Services view on the overall problem: Jih-Shyr Yih, Yew-Huey Liu: Data Center HostingServices Governance Portal and Google Map-Based Collaborations. WISE 2009: 455-462

Project: Workload Placement Analytics (2)

Migration Project Suggestion A: Migration through Image Capture Exploration

Background: Several commercial tools exist to enable migration of workloads to virtualized

environments. These tools operate at a coarse level, viewing the “workload” as essentially a

system image. For example

– VMware’s VCenter Converter aids physical to virtual migration but does this at an image

level. The VMware VCenter Converter user does not have the opportunity to selectively

migrate only those resources they need to the virtualized environment.

– PlateSpin Migrate is another migration tool that operates at the image level for its “move

workload” and “copy workload” features. PlateSpin Migrate provides a “Server synch”

feature, which synchronizes deltas between an original and a target “workload”.

– Double-Take Move provides slightly finer granularity than VMware VCenter Converter in

that it allows users the option of migrating either an entire server image or just data to

the virtualized environment.

Task: Choose a couple of image-capture tools.

1.Try a simple P2V using one or more commercial or open-source tools.

2.Define strength and deficiencies of each tool

3.Prepare a comparative assessment of these tools based on your experience.

1. is experimental, 2 and 3 could be based on the literature

42

Migration Project Suggestion A: Migration through Image Capture Exploration (Continued..)

Expected results:

– Writeup of outcomes of experiments

– Use student’ chosen tools. Prepare a comparison in a simple format to discuss with mentors.

Closest lecture: Week 10, automatic migration

Reading list:

– Christopher Clark, Keir Fraser, Steven Hand, Jacob Gorm Hansen, Eric Jul, Christian Limpach, Ian Pratt, Andrew Warfield, Live Migration of Virtual Machines, 2nd Symposium on Networked Systems Design & Implementation (NSDI ’05), 2005

– Manish Sethi, Kalapriya Kannan, Narendran Sachindran, and Manish Gupta: "Rapid Deployment of SOA Solutions via Automated Image Replication and Reconfiguration" - Proceedings of the IEEE International Conference on Services Computing, Honolulu, Hawai, 2008:155-162

– Related papers corresponding to other lectures, for discussion only:

• Nikolai Joukov, B. Pfitzmann, H. V. Ramasamy, and M. V. Devarakonda, Application-Storage Discovery, The 3rd Annual Haifa Experimental Systems Conference (Systor'10), Haifa, Israel, May 2010

• Xu Chen, Ming Zhang, Z. Morley Mao, Paramvir Bahl, Automating Network Application Dependency Discovery: Experiences, Limitations, and New Solutions, 8th USENIX Symposium on OS Design and Implementation (OSDI 2008), San Diego, CA, December 2008.

Mentors:

– Harigovind Ramasamy

43

Migration Project Suggestion B: Workload Migration through Provisioning and Configuration Migration

Background: Server migration and consolidation are a key type of enterprise IT

transformation projects. In a case when source and target of migration are not the same,

the workload migration takes the following steps:

– Provision a new server based on details provided in design starting with OS

installation. Introduce software packages in a phased approach. Initiate backup

(source) and restore (target) once new system available

Task: Choose a workload that you have a full access to.

1. Define a process and migrate this workload manually to an instance on one of available

public clouds, preferable IBM SmartCloud.

2. Understand available clouds/images and ensure the best target cloud/image selection

for the chosen workload to minimize the overall time/cost of transformation if there is

no perfect match with the source workload.

1. is experimental, 2. could be based on either experience or literature

44

Migration Project Suggestion B: Workload Migration through Provisioning and Configuration Migration (Continued…)

Expected results:

– Writeup of outcomes of experiments

– Use student’ chosen application. Prepare a process definition in format of student’s choice to discuss with mentors.

Closest lecture: Week 10, automatic migration

Reading list:

– C. Ward, N. Aravamundan, K. Bhattacharya, K. Cheng, R.Filepp,R.Kearney, B. Peterson, L. Shwartz, C. C. Young, “Workload Migration into Clouds – Challenges, Experiences, Opportunities” Proceedings of the 2010 IEEE International Conference on Cloud Computing (CLOUD 2010) :164-171

– Liang Liu, Li Ying, Qian Ma, Ke Wei Sun, Ying Chen, Hao Wang, “Automatic Model-Based Service Hosting Environment Migration”, IBM Research Report RC24437, November 27, 2007.

– Torchiano, M.; Di Penta, M.; Ricca, F.; De Lucia, A.; Lanubile, F. Software Migration Projects in Italian Industry: Preliminary Results from a State of the Practice Survey, 23rd IEEE/ACM International Conference on Automated Software Engineering - Workshops, 2008:35-42

– Related papers corresponding to other lectures, for discussion only:

• Xu Chen, Ming Zhang, Z. Morley Mao, Paramvir Bahl, Automating Network Application Dependency Discovery: Experiences, Limitations, and New Solutions, 8th USENIX Symposium on OS Design and Implementation (OSDI 2008), San Diego, CA, December 2008.

Mentors:

– John Rofrano

45

Cloud Networking: Cloud Customer Self-Help Portal for Virtual Security Zones

Motivation:– Today, Infrastructure-as-a-Service (IaaS) providers give “blank-slate” VMs to customers; configuring

those VMs is largely the customers’ responsibility

Goal– Create a self-help portal for customers to define network security policies– Develop automation capabilities for effecting customer-defined security policies

You will learn:– Enterprise/Cloud networking – Workload connectivity and Cloud security requirements– Network configuration analytics and automation

Expected results:– A cloud self-help portal for network configuration automation that enables customers to define their own

flow-control policies between their VMs and effect those policies

Grading:– At least two different VM platforms (e.g., one Windows and one Linux) must be handled– Policy/zone conflicts must be handled – Web front-end must

• Be intuitive, simple, and easy-to-use • Include per-customer authentication• Be integrated with a mock “provisioning” application that gives updated information of customer

VMs– Steady-state changes to policies or virtual zone memberships must be accounted for– Mid-term and final demonstrations expected along with a project report– You will be compared to all other students so make sure that the amount of work and details is

appropriate

Mentor: Hari Ramasamy ([email protected])

mailto:[email protected]

Network Optimization Project: Create an OpenFlow Network

Problem: Develop a graphical tool to automatically deploy/configure a software-defined network

using OpenFlow (www.openflow.org)

Configure the network to run at least 3 different types of flows

What you will learn:

Fundamentals of configuring a “cloud-positioned” network architecture

Expected results:

Development of a “graphical” OpenFlow controller, virtual OpenFlow nodes, and load

balancing and/or firewall capability in at least one of the OpenFlow nodes

Live demonstration of how the network operates according to a user's configuration

Grading

Quality of GUI, correct and automated deployment, and analysis of traffic

performance given configuration of multiple flows

Mentor: Hari Ramasamy

47

http://www.openflow.org/

transformation to the cloud...image migration –capture an image of the server and reconstitute in...

Documents