47
CLOUD COMPUTING – Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group Department of Computer Science ETH Zurich, Switzerland www.systems.ethz.ch JISBD - 2010 1

CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

CLOUD COMPUTING –Y SUIMPACTO EN LA INFORMATICA

Gustavo AlonsoSystems Group

Department of Computer ScienceETH Zurich, Switzerland

www.systems.ethz.ch

JISBD - 20101

Page 2: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Background

ETH ZürichSystems Group

www.systems.ethz.chEnterprise Computing Center

www.ecc.ethz.chCTO of Zimory GmbH

www.zimory.com

2

Page 3: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Overview

Why cloud computingCloud computing modelsA research agendaComputer science and the cloud

3

Page 4: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

The motivation for cloud computing

4

Page 5: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Cloud computing for skeptics

I WANT YOURDATA CENTER

and everything on it

Page 6: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Why cloud computing

6

Page 7: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Cloud computing for economists

A natural step for the IT industry, moving from the secondary sector of the economy (manufacturing)to the tertiary/quaternary sector of the economy (services)and exploiting economies of scale

7

Page 8: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Service model is not what we know

Closed systems and platformsExample: Appliances, automobiles

Single providerExample: iPhones / iPads

Utility detached from technologyExample: telephony, electricity, water

8

Page 9: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Widespread impact

The cloud represents the true “disappearing computer”Mobile devices as the gateway to the cloud (= the computer)

9

Page 10: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Relevant aspects

Costs

Size

Technology

10

Page 11: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Costs• Oracle and Sun SPARC SOLARIS World Record TPC-C Performance Beats IBM's Best Results on DB2 with Power 595 Server• Oracle and Sun Publish First World Record TPC-C Benchmark using Flash Technology• Oracle OpenWorld, San Francisco, Calif. – October 11, 2009

… achieved a record-breaking 7.7 million tpmC

at $2.34/tpmC

11

Page 12: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Cost of benchmarking set up

12

Page 13: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Cost of owning the system

13

Page 14: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Size matters

Cost of bad weatherRailway companiesAirline reservations

New servicesPostal servicesFinancial services

14

Page 15: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Amadeus WorkloadPassenger-Booking Database

~ 600 GB of raw data (two years of bookings)single table, denormalized~ 50 attributes: flight-no, name, date, ..., many flags

Query Workloadup to 4000 queries / secondlatency guarantees: 2 secondstoday: only pre-canned queries allowed

Update Workloadavg. 600 updates per second (1 update per GB per sec)peak of 12000 updates per seconddata freshness guarantee: 2 seconds

15

Page 16: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Amadeus Query Examples

Simple QueriesPrint passenger list of Flight LH 4711Give me LH hon circle from Frankfurt to Delhi

Complex QueriesGive me all Heathrow passengers that need specialassistance (e.g., after terror warning)

Problems with State-of-the ArtSimple queries work only because of mat. views

multi-month project to implement new query / process

Complex queries do not work at all16

Page 17: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Technology

BEFOREPeak load provisioning

One app per box

In house facilities

AFTERPay as you go

Virtualization on multicoreCapacity on demand

17

Page 18: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Why cloud computing - III

Business casePeak provisioningCosts of large installationsNeed for consolidationComputing becoming invisible

Part 1 18

Page 19: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Cloud computing models

19

Page 20: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Types of cloud

It is important to keep in mind that there are several types of clouds:

Public clouds (services publicly available)Private clouds (services within one entity)

And several types of servicesInfrastructure as a Service (Hardware)Platform as a Service (Hardware + Basic Software)Software as a Service (Full Applications)

20

Page 21: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Public Clouds

Public clouds are a service offering by a provider that are open to the public and can be used following a contractual agreementAdvantages:

Economies of scaleRobustness through numbers

DisadvantagesLegal and security aspectsLoad separation

21

Page 22: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Private Clouds

Private clouds are virtualized resources made available as a service within one organizationAdvantages

No legal and security aspectsYour own infrastructureFast provisioning of resources

DisadvantagesLimited sharing of resourcesYour own infrastructure

22

Page 23: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Private vs. public cloudsThe most common scenario in the short and medium term is likely to be private clouds connected to public clouds for quick provisioning

Private cloud facilitates transition while maintaining control over IT infrastructure

Public cloud (multiple providers) for expansion and outsourcing

23

Page 24: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

IaaS: my hw, your sw

Infrastructure as a ServiceVirtualized hardware forming a grid

Machines, memory, CPUs, diskNetworking plus load balancersInternet access

Example IaaS: Amazon‘s EC2Example of software: AppLogic from 3Tera

IaaS

SaaS

PaaS

24

Page 25: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Cloud user (company)

Iaas configuration

Cloud provider (IaaS)

Company‘s softwareCompany‘s

software

Provider‘s hardware

Developers UsersUsers Users

25

Page 26: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

IaaS: pro and against

AdvantagesFast provisioningReduced CapExLarge scale resources for small companiesYou control the software that runsLeast vendor lock-in

DisadvantagesHardware CapEx not the dominant factorDevelopment and Maintenance costs remainModel with lowest margings for provider

26

Page 27: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

PaaS: you do the tailoring

Platform as a Service(Virtualized) Hardware resourcesBasic software

databases or programming platforms

NetworkingInternet access

Example PaaS: Google Application Engine, Microsoft Azure

IaaS

SaaS

PaaS

27

Page 28: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Cloud user (company)

PaaS configuration

Cloud provider (PaaS)

Company‘s softwareProvider‘s software

Provider‘s hardware

Developers UsersUsers Users

Provider‘s software

28

Page 29: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

PaaS: pro and against

AdvantagesPlatform software already installedSimplified maintenance of platform softwareControl of the application and dataReduced CapEx for hardware and key software

DisadvantagesSome development and maintenance costsOften inflexible in what can be done and deployedVendor lock-in at the platform level

29

Page 30: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

SaaS: you use the service

Software as a ServiceFull applicationAvailable through a web browserInternet access

Example Saas: SalesForce, Google Apps

IaaS

SaaS

PaaS

30

Page 31: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Cloud user (company)

SaaS configuration

Cloud provider (SaaS)

Provider‘s hardware

DevelopersUsers Users

Provider‘s software

Users

31

Page 32: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Saas: pro and against

AdvantagesNo CapEx or OpExCost related to useNo maintenance or developmentModel with highest margins for providerLarger software markets (no differentiation)

DisadvantagesVendor lock-in at the application levelAvailability

32

Page 33: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

A research agendaSystems Group ETH

33

Page 34: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Appliances for the cloud

34

Hardware

Application

Platform

Operating System

Data processing on FPGAs

Crescando

Multimed

Barrelfish

Access Alfredo

Page 35: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

35

CPU-based Systems

0%

20%

40%

60%

80%

100%

300'000 pkts/s 1'000'000 pkts/s

CPU-based FPGA

[Streams on Wires, VLDB ’09]

data input rate

pack

ets

proc

esse

d

CPU-based System drops UDP packets

Network-memory-CPU bottleneck

FPGA → not a single UDP packet dropped

60%

36%

100% 100%

Page 36: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Complex Event Detection at Wire Speed with FPGAs 36© ETH Zurich

Programming FPGAs

Network FPGA Notification End SystemData Stream

Complex EventQuery

Queries Compilation Synthesis → Map → P&R BitstreamVHDL

Page 37: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Complex Event Detection at Wire Speed with FPGAs 37© ETH Zurich

Optimal network utilization

Min. Ethernet frame = 64 bytesInterframe Gap = 12 bytesPreamble = 8 bytes

84 bytes (672 bits) per packet

→ 1 Gbit/672 bits = 1’488’095 packets

Maximum number of packets on a gigabit link

Our measurements1’451’373 packets/second1’451’373 packets/second x 672 bits = 975.3 Mbit/s ≈ 1 Gbit/s

Page 38: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Barrelfish

38

Page 39: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Postgres on Multicore Architectures

Nehalem EX 4 CPU Sockets 8 Cores per CPU128 GB RAM

TPC-W Browsing 24GB Database

39

• PostgreSQL can’t take advantage of the extra resources

Page 40: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Multimed

40

Query Q2 -W

HPI - SOC Lab Symposium 40

Query Q1 -R

Query Q1 -R Query Q1 -R

Query Q2 -WQuery Q3 -R

Query Q3 -R Query Q3 -R

Page 41: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Crescando on 1 Machine (N Cores)

Systems Group. D-INFK. ETH Zurich41

...

Split

Scan Thread

Scan Thread

Scan Thread

Scan Thread

Scan Thread

MergeInput Queue(Operations)

Input Queue(Operations)

Output Queue(Result Tuples)

Output Queue(Result Tuples)

Page 42: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Crescando in Data Center (N Machines)

Systems Group. D-INFK. ETH Zurich42

Page 43: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Alfredo

Page 44: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Computer science and the cloud

44

Page 45: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

New opportunities

Cross layer optimizationsSystem co-designNew paradigms (system and languages)Algorithms in hardwareHybrid compilers (HW-SW)New data processing languages

45

Page 46: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

New challenges

Computer no longer visibleLayers of abstractions blurringWhat is a program?What does a program do?Business models changingEngineering skills in high demand

46

Page 47: CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Computing-JISDB … · CLOUD COMPUTING –Y SU IMPACTO EN LA INFORMATICA Gustavo Alonso Systems Group. Department of Computer Science

Conclusions

Three Chinese curses1. May you live in interesting times

2. May the government be aware of you

3. May you find what you are looking for

47