Socieconomics applied to Distributed Systems-1.ppt

44
8/9/2019 Socieconomics applied to Distributed Systems-1.ppt http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 1/44 Socieconomics applied to Distributed Systems OR When Tamas met Dave

Transcript of Socieconomics applied to Distributed Systems-1.ppt

Page 1: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 1/44

Socieconomics applied to

Distributed Systems

OR

When Tamas met Dave

Page 2: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 2/44

OR

Page 3: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 3/44

Page 4: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 4/44

Goals

•  To !tremely brie"y #and presumptuously$capture the ist of a part of what you havebeen discussin in this class% #&hD studentstake note' Tortured (uali)cations alore$

•  That is Dave*s stu+ #to which will add mine$#models-$

• .ink it to what Tamas has been teachin#some-$ of you #bi data$

• How both are needed for the future/

Page 5: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 5/44

Page 6: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 6/44

4y &hD 5ears

• Was primed due to my backround fordoin the kind of work that did% 3nd itwas 6ust sheer ood luck that Dave

ended up bein my roommate%

• Obsessed with fairness and e7ualityand feelin pain at the way thins were#esp% vi8 a vi8 the 9economic system:%

Page 7: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 7/44

Rewardin 2+ort in &;&Systems

•  Took this idea from &articipatory

economics #this is where randmacomes in% Grandma is me, amonother people$

• 3 proposed economic vision for amore 6ust and e7uitable future%

Page 8: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 8/44

•  The followin few slides are from a presentation ave for this work%

• ts not merely la8iness on my part%

• do want to show the 9political: stu+ that ended

sayin at a so<called technical conference=and infact had a )ht with my supervisor over it%

• 1ut as you will see, the point was that the political,social economic #call them whatever$ thins are

already there' 2>istin in how current systems hadbeen desined% They are 6ust taken as conventionalwisdom

• 3nother point bein, feel free to piss people o+/ #not

too much please$

Page 9: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 9/44

  3bout this work

 The motivation for this work comesfrom considerin allocation of resourcesand wealth distribution in the world=

t seems that the way thins currentlywork lead to thins such as=%

 The recent 2conomic crisis-

Well, no/ Or at least not 6ust%

Page 10: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 10/44

  0ris2S all over

Over 50 % of the world population lives on less than

?;@day

A people own more wealth than BC countries

2tc, etc, etc and etcetera

It perhaps takes diferent glasses to see all thesecrises rather than just one crisis on which the ocus isright now:

Page 11: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 11/44

 52S, 02 G.3SS2S,D22D, T H3S 3.. GO2 1OE2RS/

Page 12: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 12/44

One comes across the followin all the time in the&;& literature%

ncentives, rationality, fairness, free ridin,eFciency etc=which is all hunky dory%

1ut one would have thouht if its all so bad in thereal world, people in &;& would look not only atthe mainstream but also at alternatives%

4aybe e>plore some other economic systems thathave been proposed--

Well, this work ives it a shot=

  .inks to &;&--

Page 13: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 13/44

What do we propose-

3 new incentive system for &;& systems inspired from&articipatory

2conomics #&arecon$ #see parecon%or for details$

Reward 2+ort instead of contribution

2+ort- Doin the best one can relative to capacity

1ut why e+ort- 4ainly to increase distributive fairness% &eers have

Di+erent capacities to contribute% Some simply cannot contributea

lot%

3nd also increase social welfare' sum of all peer*s utilities

o, but you may still ask, WH5- WH5 o WH5-

Page 14: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 14/44

So what is the problem with current allocation in &;&

systems%

or that we need to consider' What is desired of anysystem that allocates resources for production and

consumption-

OK!

Page 15: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 15/44

0ooperation

1asically the oal of all incentives schemes% Giveyour resources to the system%

Received wisdom is that people won*t cooperateunless incentivi8ed to do so%

Page 16: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 16/44

2Fciency

1asically means resources shouldn*tbe wasted% So for e>amplethousands of people homeless and

thousands of hotel rooms empty iswastefulness%

n &;&, minimally, upload resourcesshould be utili8ed and not wasted%

Page 17: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 17/44

How is eFciencymeasured-

ormally economists, and also &;&desiners, talk about &areto optimality%

o one can improve their lot without

makin someone worse o+%

ery ood but not necessarily fair

2%% start by ivin all resources to one

person and nothin to the rest and youhave &areto optimality%

Page 18: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 18/44

2Fciency criterion

Given the inade7uacy of &areto optimality,economists have used a concept called the

Ecienc !riterion: f the overall bene)ts toany and all people of doin somethinoutweih the overall costs to any and allpeople, it is eFcient to do it, and vice versa%

3ain, who is to say which 9any: people, whichroup-

3lso, it*s deemed that'2Fcient outcomes  Social Welfare

Page 19: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 19/44

R IJI

G K

1 CC

R ILA

G IMC

1 IJ

R K

G BA

1 MJ

R K

G IBL

1 ILI

R I;A

G IJK

1 ;KI

R IIM

GIM;

1 ;IN

  The fact is value 6udments are implicit inthe

eFciency criterion%

3 desiner has to make value 6udments

on what is a better solution% avor fast-avor

slow- avor somethin else-

n our paper, we desire eFcient

outcomes, butsuch that they are fair and e7uitable to

the less

resourceful peers in the system%

Page 20: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 20/44

t could be arued that it is fair that the fast peers, whocontribute more to the system in terms of volume,

overall et better service and more rewards from the

system% However, this de)nition of fairness assumes

ma>ims of remuneration that reward peers for their fast

connections%

 There are two familiar ma>ims of remuneration'

a$ &ayment accordin to value of one*s personal

contribution and one*s productive property

b$ &ayment accordin to value of one*s personal

contribution only

Page 21: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 21/44

nspired from &articipatory 2conomics #&arecon$ we use anovel ma>im of remuneration'

c$ &ayment accordin to e+ort

We applied this to a$ 1itTorrent and b$ 3 credit based

sharin ratio enforcement scheme

Page 22: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 22/44

ow=

3pplied it to 1itTorrent% 3t that time,didn*t know anythin about 1itTorrentprotocol% 3 smarter friend, 4ichel4eulpolder had made a lihtweiht1T simulator in which he interatede+ort based incentives

• still remember 4ichel came runnin

to my room and showed me thisraph

Page 23: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 23/44

Page 24: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 24/44

3nd my reaction was=

Why are you SO happy-

• 3 wron reaction but not too much%

1ut what am tryin to say here-

• When you apply such your ideas, you usuallyknow the results you are oin to et% What*sinterestin is when you don*t et those results%

• n this particular case, was wron, because itsnot straihtforward that rewardin accordin toe+ort is better%

Page 25: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 25/44

Grandma 2+ect Works/

• So mentioned &areto% The fact is that when brouht this up in Delft, we #includin me$ were allconfused about this

• We didn*t know what was what% Really% am notkiddin% o one knew this stu+% 3nd its really basic

stu+%%%

• 1ut randma e+ect works because not only did thispaper end up ettin published, it has been cited afew times #as much as cheat 1T$ and more

importantly, was 6ust checkin, it has been used toapply e+ort based incentives in other settins, suchas collaborative 0& slot sharin etc%

 • &oint bein-

Page 26: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 26/44

3nother salvo/

• Desin Space 3nalysis for 4odellinncentives in Distributed Systems

• 3n alternative to ame<theoreticanalysis of distributed protocols

Page 27: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 27/44

 Traditional vs Our 3pproach

Specif y

Desi

nSpace

Simulation1ased

3nalysis

&redicted

Outcomes

  4any

&rotocols

1rainstorm@ntuitio

n

Game Theoretic3nalysis

&redictedOutcomes

&rotocolariant

Page 28: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 28/44

Page 29: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 29/44

3nd now 1i Data/

• So what Dave has been talkin about inthis course is reat

• 1ut, current trends seem to be movin

away from that• Social Science or randma social

science is still bein talked about but ina di+erent way all toether%

ow it seems the randma doesn*tneed to know anythin apart fromstatistics #and some machine learnin$

• Or 3esop .earns Statistics

Page 30: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 30/44

ow the idea is

• We have all this 9data: riht- &eople leavintheir )nerprints

sin this we can follow how people interactwith each other and the environment

• 3nd then usin loads and loads of data, we

can describe mathematical connectionsbetween information #ideas$ "ow andhuman behavior/

2 l

Page 31: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 31/44

2>ample• n his book 9Social &hysics' How ood ideas spread=: 3le> &entland, #one of$

the Godfather#s$ of the bi data movement, ives this e>ample of a bi data

success

•   Traders share tips on a social network% Sometimes only a few traders make alot of pro)t% 4any solutions have been tried to stop this%

•  His team found #by analy8in the 4..OS of detailed messae amon traderson a social network$ they discovered herdin and the e+ects of social in"uence,

in which the traders overreacted to each other%

•  ow this was always known wasn*t it- 1y randma and her friends, Tom, Dickand Harry

 

Page 32: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 32/44

Other e>amples•

One study of a social network they did foundthat individual who adopted an eneretic,enain interaction style that created moreinteractive conversations ended up bein

more important to idea "ow in the network%%• Hmm%%

• 1ut before we make 9fun: of them

Remember the Schellin 4odel that youstudied some weeks back% Grandma and3esop are everywhere and not 6ust in bidata

Page 33: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 33/44

So let*s critici8e or praise1i Data from three anles'

• 4ethodoloy

• Results

• Goals

4 th d l

Page 34: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 34/44

4ethodoloy•  The idea is usually if you read such papers,

we collected billions of messaes frommillions of users=

• Why is it necessarily better to study thins

at such a rand scale-

• 3n imperfect analoy% maine if you want

to hear what*s bein said in a crowdedroom% The thrust of the conversation%

• oise

0ollect data for the heck of

Page 35: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 35/44

0ollect data for the heck ofit/

• 0ollect every facet and dimension of behavior,

communication, and social interaction amonits members% Do that for lon periods

3nd-

• Sub6ect this data to statistical analysis

• Have to know what to look for% #Smart onesprobably do$

Page 36: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 36/44

Page 37: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 37/44

Results #;$ The 1ad

• &rediction based results #related tooals$%

Goole predicted epidemics based onminin search trends%

• 3lso ot it wron

.ookin at black bo>es inside cars to setinsurance policies #-$ for e>ample

• &rediction is a tricky business esp%, insocial a+airs, let alone the hard science%

Page 38: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 38/44

Page 39: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 39/44

Goals• 1y discoverin how people behave and

analy8in patterns of behavior #worthy e+ort$,we are discoverin why some thins mihthappenPcrashes, revolutions, bubbles etc

• OE, randma and 3esop also had theirinsihts, but presumably studyin thesethins on a rander scale, would hopefullylead to more insihts% urther, Grandma and3esop can be wron on some issues%

Page 40: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 40/44

Goals #;$

• 1ut based on the above, #asmentioned in the results$, theultimate oal is to shape outcomes%

Set policies for rowth andinnovation, for eradication of povertyand warPin short, 1i Data topia/

•  This is danerous/

f you want a more detailed

Page 41: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 41/44

f you want a more detaileddiscussionof the model vs

bi data approach•  The most, and perhaps the only, hih pro)le

9debate: on this topic has been

S

  &eter orvi 

Page 42: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 42/44

Page 43: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 43/44

 Tamas 0ourse

• 4ost 6obs in computer science #for &hDholders at least$ these days are for datascientists% 3ll kinds of companies, from2lectronic 3rts #yes, that ames company$to 1ookin%com #remember-$ are lookin for

data scientsits

• So please do yourselves a favor and take Tamas course%

•  This is not an e>aeration' as far as ettina 6ob is concerned it could be one of themost important courses that you ever take/

Page 44: Socieconomics applied to Distributed Systems-1.ppt

8/9/2019 Socieconomics applied to Distributed Systems-1.ppt

http://slidepdf.com/reader/full/socieconomics-applied-to-distributed-systems-1ppt 44/44

3nd on that note

 Thank you for listenin, and thanksto Dave and Tamas