If w p f fxtsnz En l list then added to s If x F Lto 4 Ee - - PDF document

if
SMART_READER_LITE
LIVE PREVIEW

If w p f fxtsnz En l list then added to s If x F Lto 4 Ee - - PDF document

Hitters Heavy stream of Its e g IP addresses a ay ay 64 each 2 n c ex ai n For any ten true t f let ett tires in appears 9,92 at a ffrequency Goal f allelts Fa return set x large using of t space fxt Tz heavy


slide-1
SLIDE 1

Heavy Hitters

stream of Its

a

ay

ay

e g

IP addresses 64

each

ai

c

n

ex

n

2

For any true

ten

let

f

t

tires

ett appears

in

9,92

a

at

ffrequency

Goal

return

allelts

x

set

f large

Fa

using

  • f

t

space

If

fxt Tz

add

to

list

  • f

heavyhalters

Provably impossible

with

sub linear space Modified goal

If

fx

Yz

add it

to list

If

x

added to

list

then

w p s

l

f

fxtsnz En

Ee

Suppose

4

20

E

0.01

F Lto

If

fx

2

x

is definitely

  • n list of HH

0.08N

if

x added to

list then with prob 31 Lto

f

Yz

En

  • .O 5h
  • 01h

O 04h

Modified goal

If

fx

Yz

add it

to

list

If

x

added to

list

then

w p s

l

f

fxtsnz

Engg

use elegant verysimple data stmche

Count MinSketch

des

l

I

s

e

slide-2
SLIDE 2

Count Min Sketch designer specifies

K

  • e

bye

keep

2D

array

i

l

hash tables eachof size b

b

4YIIsepae

hashtable

i

no

i

i

e

5

he

e

when

ett

shows

up

IncCx

tf

is

jel

increment

CMS j

hj xD

Observation

at any

tinet

  • mg

hjk

fit

V j

tx

County

return

min

CmsCj

hjlxD

is's

El

if Countfx Fc

by ObsuralnLO

Couth

add

to HH list

Assump

hasi

hashfus

behave like

random maps in thefollowing sense

Vxfgyc LDV

lejelprfhjlxfh.ly

tb

hashas

I

El

are indep geochotm

Analysis

Fix the t elt

X

Define

Zj

2j3f

zj

fxttff.gg

Yq

w.j q.e

hjcxi hjlD

slide-3
SLIDE 3

is

  • aw

E 2j

ftty fytEwxy

e fxtttbefxtth

bprfhjcxj

hg.ly

ten

t

b

E 2J fxt

IF

MaskaisInegu

By

Markov'sInga

2 2

X

is

anonnegative

n v

Prix

xECxD

et

p

L

Prf2J Ext 2

at

castzo

PY.FI 3fxtt2I

yeuntzbEl

2e

Prf2,3fxtt2f

223k

25

Ze fxtt

Pr

2

fxtt2f Prey fifty

Prue fxtt2Ib

indef ghashby

by

I

7 Mdgsof2J's

Pr

County

s fxtt 2mg

E

1

min

Z

W

2h

I ee

I

En

T

I

K

E

  • r

Modified goal

If

f

Yz

add it to list

If

x addedto list

then wp s l f fxtsnz

Engg

8 122

En

2hbE

l eogff b

2zf

slide-4
SLIDE 4

6

200

e

5

E Fo

  • .O I

b

5

error prob

is

432

I

004N

if

has f

so 04h

definitely

  • utput

ifontputx thnwitnprob 3152 f 37z En

p

primenumber

n 004N

0.0in

to 03in

H

hcxt fextgfmodpmodbffzee.IT

g

H1

pcp c

universal

family

next

t.is mdPmdbI

93

If

h

is chosen uniformlyatranderm

from Jt

V x fyprfhc.no

hcg

bt

T

heft

slide-5
SLIDE 5

Markov's Inequality

X nonnegative

n v

For any

positivecast

c

Pr X

CEH

E

z

Prog

F G

15

t.EE H

Z

O

t

EcECX

Pr

X

x

X CECX

cECx

Pr X CECH

pr

XxcECx

tf