+ All Categories
Home > Documents > Bogdane BUTNARU, Benjamin NGUYEN, Georges GARDARIN, Laurent YEH PRISM LAB. UNIVERSITY OF VERSAILLES,...

Bogdane BUTNARU, Benjamin NGUYEN, Georges GARDARIN, Laurent YEH PRISM LAB. UNIVERSITY OF VERSAILLES,...

Date post: 13-Dec-2015
Category:
Upload: edwina-cole
View: 216 times
Download: 0 times
Share this document with a friend
Popular Tags:
7
Bogdane BUTNARU, Benjamin NGUYEN, Georges GARDARIN, Laurent YEH PRISM LAB. UNIVERSITY OF VERSAILLES, FRANCE Namur 20-23 Octobre 1 XQ2P: Efficient P2P Processing of Time Series XQueries
Transcript

Bogdane BUTNARU, Benjamin NGUYEN, Georges GARDARIN, Laurent YEH

PRISM LAB.UNIVERSITY OF VERSAILLES, FRANCE

Namur 20-23 Octobre

1

XQ2P: Efficient P2P Processing of Time Series XQueries

BDA 2009- Demo

Objectives

Namur 20-23 Octobre

2

Objectives: Propose a simple model for the management of

large and multiple time series (TS) in XML Demonstrate XQ2P, our prototype for efficient

P2P processing of XQuery TS Introduce an application in the context of stock

analysis with multiple and large TS (>1K * 1M values).

BDA 2009- Demo

Basic Data Model (1)

Namur 20-23 Octobre

3

• Derived from the RoSeS project model• Based on a vector space of TS with potentially

infinite items

• Each vector is associated with a calendar giving each point in time ; calendars can be shared between TSs

• There exists two possible and distinct null values:• the empty (or non-exist) value (denoted “!”) • the unknown value (“?”)• Null values can be combined :

BDA 2009- Demo

Basic Data Model (2)

Namur 20-23 Octobre

4

The vector space is equipped with relational-like operations mapping one, two or more TS to one.

It includes window-based operations similar to those proposed in the current XQuery 1.1 WD.

BDA 2009- Demo

Stock Selection and Strategy Evaluation

Namur 20-23 Octobre

5

Stock technical analysis attempts to consider stock prices and volumes as temporal signals

It analyses these signals based on indicators, patterns, or events.

Sliding window operators are often used

A generic window-based operator: WINfun(S,w) = {[t, val] | val = fun([t-1, val1], [t-2,val2], … [t-w, valw ])}. Cost of computing the whole TS per window is polynomial

w

tswtStVw

w

tStV

t

wt

][][]1[*1][][

6

Develop XQuery 1.1 functions, using the window clause

P2P : Chord DHT-based index :

put(key=<name>, content=(<peerId><start><end>)*).

Each peer has a main memory cache with a replacement policy (e.g., FIFO)

Loading a base TS segment or producing a derived one keeps the series segment in memory cache.Namur 20-23 Octobre

BDA 2009- Demo

Experimental Evaluation

Namur 20-23 Octobre

7

Qizx 3.0 XQ2P (on 1 peer) XQ2P (on 128 peers)

GB Network Load Time

624ms 624ms 624ms + 1140ms routing

Bandwidth 78MB 78MB 78MB+1MB routing

Disk Load Time 1986ms 14370ms 113ms

WAVG100 comp. 99 042ms 1 092ms 1140ms routing + 128ms computation

MACD100 comp. 375 000ms(using avg)

54 819ms 1140ms routing + 580ms comp.

+ 624ms netw.


Recommended