Distributed Systems: What?

Why? And bit of How?

Srinath Perera

I cannot cover Distributed Systems in 30 minutes!

But, I can tell why you might want to learn Distributed

Systems in 30 minutes!

http://www.flickr.com/photos/uwehermann/82753155/sizes/m/in/photostream/ and http://www.flickr.com/photos/peterpearson/5921765552, licensed under CC

What is a Distributed System?

"A distributed system is one on which I cannot get

any work done because some machine I have

never heard of has crashed.“

--Leslie Lamport

What is a Distributed System?

“A system in which hardware or software components located at networked computers communicate and coordinate their actions only by message passing.” - [Coulouris]

“A distributed system is a collection of independent computers that appear to the users of the system as a single coherent system.” - [Tanenbaum]

Characteristics and Challenges• Fault

Tolerance• Scale• Transparency

• No Global Clock• Communication

only by message Passing

• No Global State• Independent

Failures

Photo by John Trainoron Flickr http://www.flickr.com/photos/trainor/2902023575/, Licensed under CC

Fallacies of Distributed Systems

• The network is reliable.• Latency is zero.• Bandwidth is infinite.• The network is secure.• Topology doesn't change.

• There is one administrator.

• Transport cost is zero.• The network is

homogeneous.http://www.flickr.com/photos/12587661@N06/2300406685, @Michael Gwyther-Jones, Licenced CC

Why Distributed Systems

• Need to build bigger systems• Many usecases are inherently distributed • To avoid failures • Omnipresence

– if you buy food from a super market – If you buy a book from a Bookshop Chain– If you search in the Web– If you use a GPS navigator – If you turn on your My 10 list– If you pay a bill– If you use your mobile App

A System Usecase Classification

• Processing Data (Moving vs. Stored Data)

• Servers: Receive, Process, and Respond

• Running User provided Jobs

• Data Storages and Provenance

http://www.flickr.com/photos/kelsea-groves/5535666329/

Usecase: Processing Data: React to Sensors

• Many sensors: Weather, Travel, Traffic, Surveillance, Stock exchange, Smart Grid, Production line

• Monitor, understand, and react to events• Usually handled with CEP (e.g. Esper, Stream Base, Siddhi) or Stream

Processing (S4, Twitter Stream)http://www.flickr.com/photos/imuttoo/4257813689/ by Ian Muttoo, http://www.flickr.com/photos/eastcapital/4554220770/,

http://www.flickr.com/photos/patdavid/4619331472/ by Pat David copyright CC

Usecase: Processing Data: Target Marketing

• Receive data about users continuously: e.g. web clicks, what they brought, what they liked and do not like, what their friends like and brought

• Build models, index information in the background • Send him advertisements that best matches his preferences

– have to do this quickly– in few (say 50) milliseconds

• Cloud be the next billion dollar problem

Usecase: Receive, Process, and Respond: Online Store (e.g. Amazon)

• Many Sellers selling many items and Many Byers

• List of all items, with their specs

• Index items by many dimensions and support search

• Support checkout, track the delivery, returns, ratings, and complains

• Supported by partitioning sellers/ items across many nodes

Usecase: Running User Provided Jobs : SETI@Home

• Many people volunteer their computing power

• Scientists submit computing jobs to the system

• Broker and match resources with jobs, run them and return results. Handle failures. Avoid free riding.

• Considered biggest computer in earth (505 TFLOPS, 150k active computers)

Finding EThttp://www.elfwood.com/~axthony/Staring-Aliens.2552052.html, Licensed CC

Usecase: Data Storages and Provenance (Sky Server)

• Telescopes (Square Kilometer Array) keep collecting data from the sky (Tera bytes per day)

• Sky Server let scientists to come and see the sky of a given location, as seen at a given time.

• Moving data takes long time. 1TB takes– 100 Mbps network : 30 hrs– 1 Gbps network : 3 hrs– 10 Gbps network : 20 minutes

• Given a data item, need to track how it is created, equipment accuracy, transformations used etc.

http://www.fotopedia.com/items/flickr-518876976 and http://www.geograph.org.uk/photo/103069, Licensed CC

Mobile Sensor Crowdsourcing• Mobile phones are now like a

weather center: has – a barometer– temperature sensor– proximity sensor– GPS– moisture sensor

• Get volunteer phones to send sensor data (Crowd source).– report on weather– crop diseases (agriculture

officials)– epidemics (from hospitals,

doctors) • Use that to do weather

predications, crop disease and epidemic spread

• Moving Sensors (Polar Grid)

http://www.fotopedia.com/items/flickr-2548697541 , http://www.geograph.org.uk/photo/1534209, and http

://www.yourbdnews.com/2011/10/17/samsung-files-to-halt-iphone-4s-in-japan-australia/iphone-

4s, Licensed CC

Mobile: Solving the Last Mile Problem

Great! lets see what Distributed System

technologies have made these use cases possible!!

Distributed Systems Timeline/HistoryPeriod Topics

1965-late 70s Parallel Programming, Self Stabilization, Fault Tolerance, ER Model/ Transactions, Time Clock

1980s Consensus and impossibility, SQL, Distributed Snapshots, Replications, Group Communication

Early 90s Linearizability, Parallel DB, transactional Memory, RAID, MPI

Late 90s Volunteer Computing, P2P file sharing, Complex event processing

Early 2000 Oceanostore, Web Services, Symantec Web, REST, DHT, Pub/Sub, Grid, Autonomic Computing, Google File System, Virtualization, SOA, Map reduce

2005-2010 Cloud, NoSQL, Mobile Apps, Data Provenance

Theoretical Computer Science• Concerns with – Coordination algorithms: Leader

Election, multi-cast, distributed locks, barriers, snapshot algorithms

– Impossibility results, upper and lower bounds

– Distributed versions of some centralized algorithms (e.g. shortest path)

– Lot of work done on 70s, and layed the ground work for Distributed Systemshttp://www.flickr.com/photos/lodz_na_nowo/5690492370/

http://xkcd.com/384/ http://www.flickr.com/photos/quinnanya/4990131194/sizes/z/in/photostream/

, Licensed CC

Communication Protocols

• Request/Response– RMI, CORBA, REST/HTTP,

WS, Thrift• Publish/Subscribe• Distributed Queues• DHT (Distributed Hash

Tables)• Gossip/ Epidemic

Protocols • Whiteboards

http://www.flickr.com/photos/novecentino/2596898279/, Licensed CC

Message

Request/Response and Architectural Styles

• Message formats • RMI, CORBA, REST/HTTP, Web Service, Thrift

• Architectural Styles – Remote Procedure Calls (RPC)– Distributed Objects – Service Oriented Architecture (SOA)– Resource Oriented Architecture (ROA)

Known Distributed Architecture Patterns

• LB + Shared nothing Nodes • LB + Stateless Nodes + Scalable

Storage • DHT (Distributed Hash Table)• Distributed queues • Publish/Subscribe Broker Network • Gossip architectures + biology

inspired algorithms• Map reduce/ data flows • Stream processing • Tree of responsibility

LB + Shared Nothing and 3-Tier

• Most common scaling pattern• Most architectures follows this model

Storages• Single Database • Replicated Databases • Parallel Databases (Sharding)• NewSQL (In-Memory,

sharding .. Highly optimized)• NoSQL (Column Family, Key

Value pair, Document)

OracleOf Delphi

Building Scalable Systems• Single Machine• Shared Memory

Model • Clustering (State

Replication through group communication)

• Shard Nothing • Loose Consistency

with Shared nothing

http://www.fotopedia.com/items/louromig-8P4w6xtSgbY, Licensed CC

Publish Subscribe and EDA

• Many publishers send events• Subscribers register events, and a publish/subscribe

network match and redirect events • Have scalable implementations • Basis for event driven architectures

Cloud Computing• Ability to buy computations power,

storage, or execution services as an Utility, on demand.

• Best way to explain it is by comparing it to Electricity

• Idea is a big pool of servers and share. • Economics of scale through Optimize

large scale operations.• Resource Pooling. • No need for capacity planning, start

small and grow as needed. • Outsource and enabling specialization.

photo by LoopZilla on Flickr, http://www.flickr.com/photos/loopzilla/2328231843/sizes/m/in/photostre

am/, Licensed under CC

Where do go from here?

If You Plan to Learn about Distributed Systems

• One of the fields to learn by doing • You have to be a good

programmer– a patient one (Debugging)– Lazy one (but intelligent)

• Start by writing some Web Services, request response stuff

• Stop reinventing the wheel, start using tools (middleware)

• Learn Zookeeper • Take a class – read, write code,

debug, .. http://www.flickr.com/photos/mariachily/5250487136, Licensed CC

Distributed System Community• Based around ACM, IEEE, and USENIX• Well known journals

– IBM System journal, ACM Operating Systems Review, ACM Transactions on Computer Systems, IEEE Distributed Systems Online, IEEE Transactions on Parallel and Distributed Systems

• Conferences– Theory: ICDCS, SPDC– SOA/Cloud : ICWS – E-Science, Parallel Programming : HPDC, SC, E-

Science, Ccgrid– Systems : USENIX, Middleware, ACM Symposium on

Operating Systems Principles, FAST, LISA, OSDI– DB : Sigmoid record, VLDB

• Awards – Turing Award – Edsger W. Dijkstra Prize in Distributed Computing

http://www.flickr.com/photos/dullhunk/4187914071, http://www.fotopedia.com/items/flickr-1544709148, Licensed CC

Few Must Read Papers• System Structure for Software Fault Tolerance (1975)• Reaching Agreement in the Presence of Faults (1980)• Time, Clocks, and the Ordering of Events in a Distributed System (1978)• Reaching agreement in the presence of faults(1980) and The Byzantine

generals problem” (1982), • End-to-End Arguments in System Design (1984)• A Note on Distributed Computing (1994)• Scale in Distributed Systems, (1994)• The Google File System (2003)• Chord: A Scalable Peer-to-peer Lookup Service for Internet Applications,

(2001)• The Google file system (2003)• Xen and the Art of virtualization (2003)• MapReduce: Simplified Data Processing on Large Clusters (2004)

Some Open Challenges

• Every thing Data: Analytics, AI, Data Mining (Distributed versions of many algorithms)

• Complex Event Processing (CEP)• How to Scale?• Middleware for the Cloud• Scalable Storage • Provenance• Workflows• Guard against DDoS and other

Distributed Security Issueshttp://www.flickr.com/photos/brianscott/5474210001,

Licensed CC

Questions?

Copyright by romainguy, and licensed for reuse under CC License http://www.flickr.com/photos/romainguy/249370084