Lecture: Trends and Concepts in the

Software Industry

Introduction

Dr.-Ing. Jrgen Mller

Course Statistics h#p://www.openHPI.de 13,126 par8cipants, 4.161 ac8ve

contributors in discussions 6 course modules for 6 weeks 35 individual module items (videos /

slides / reading material) 300 exam / assignment / self-test

ques8ons 309 slides explicitly created for this course 215 pages book reading material 8h cut video material (12h raw material)

Teaching Team: 4 colleagues

2

0"20"40"60"80"100"

0"

1000"

2000"

3000"

4000"

Week"1"

Week"2"

Week"3"

Week"4"

Week"5"

Week"6"Exam"

Number"of"Par8cipants"in"Assignments"&"Exam"Average"Results"in"%"

Goals

Deep technical understanding of a column-oriented dic8onary-encoded in-memory database and its applica8on in enterprise compu8ng

The future of enterprise compu8ng Founda8ons of database storage techniques In-memory database operators Advanced database storage techniques Founda8ons for a new enterprise applica8on development era

3

December 2006: The Basic Idea

4

Learning Map

Founda(ons for a New Enterprise Applica(on

Development Era

Founda(ons of Database Storage

Techniques

The Future of Enterprise Compu(ng

Advanced Database Storage Tech-niques

In-Memory Database Operators

Chapter 1: The Future of

Enterprise Computing

Dr.-Ing. Jrgen Mller

New Requirements for Enterprise Computing

Sensors

Tracing Pharmaceu(cal Packages in Europe 15 billion packages / 34 billion read events per year Distributed repositories for storing read events References to read events are stored in central discovery service

Monitoring F1 car Performance Between 300 and 600 sensors per car Mul8ple events per second per sensor Tracking every Grand Prix lap or test run

8

Airplane Maintenance at Boeing Maintenance workers at Boeing write reports aYer repairs Reports and men8oned part numbers get indexed Analy8cs reveal which parts in other planes may be defec8ve

Medical Diagnosis at Charit Doctors write medical reports aYer every diagnosis Diagnosis and men8oned symptoms get indexed Comparison with similar cases for op8mal treatment

Combination of Structured and Unstructured Data

9

Mobile Mobile inverts tradi(onal corporate structures Enables customer-facing

personnel

Response 8mes < 1 second

Example: Dunning

Sales, Service

Opera8ons

Controlling

Consolida8on

Strategy

Consumers / Customers

CEO

10

Learning Map

Founda(ons for a New Enterprise Applica(on

Development Era

Founda(ons of Database Storage

Techniques

The Future of Enterprise Compu(ng

Advanced Database Storage Tech-niques

In-Memory Database Operators

Enterprise Application Characteristics

Dr.-Ing. Jrgen Mller

Transac(onal Data Entry Sources: Machines, Transac8onal Apps, User Interac8on, etc.

Text Analy(cs, Unstructured Data Sources: web, social, logs, support systems, etc.

Event Processing, Stream Data Sources: machines, sensors, high volume systems

Real-(me Analy(cs, Structured Data Sources: Repor8ng, Classical Analy8cs, planning, simula8on

Data Management

CPUs (mul*-Core + Cache)

Main Memory

Challenge: Diverse Applications

13

Modern enterprise resource planning (ERP) systems are challenged by mixed workloads, including OLAP-style queries. For example: OLTP-style: create sales order, invoice, accoun8ng documents,

display customer master data or sales order OLAP-style: dunning, available-to-promise, cross selling,

opera8onal repor8ng (list open sales orders)

But: Todays data management systems are op8mized either for daily transac(onal or analy(cal workloads storing their data along rows or columns

Online Transac8on Processing

Online Analy8cal Processing

OLTP vs. OLAP

14

Drawbacks of the Separation

OLAP system does not have the latest data

OLAP system does only have predened subset of the data

Cost-intensive ETL process has to synch both systems

There is a lot of redundancy

Dierent data schemas introduce complexity for applica8ons combining sources

15

OLTP Access Pattern Myth

Workload analysis of enterprise systems shows: OLTP and OLAP workloads are not that dierent

16

Combine OLTP and OLAP data using modern hardware and database systems

to create a single source of truth, enable real-(me analy(cs and

simplify applica8ons and database structures.

Addi8onally, extrac8on, transforma8on, and loading (ETL) processes pre-computed aggregates and materialized views become

obsolete.

Vision

17

Many columns are not used even once

Many columns have a low cardinality of values

NULL values/default values are dominant

Sparse distribu8on facilitates high compression

Standard enterprise soYware data is sparse and wide

Enterprise Data Characteristics

18

Low Cardinality of Values Within Many Columns

Results from analyzing nancials

Dis8nct values in accoun8ng document headers (99 a#ributes)

CPG Logis8cs

High tech Discrete manufacturing

Banking

19

Many Columns are not Used Even Once

55% unused columns per company in average 40% unused columns across all companies

0%10%20%30%40%50%60%70%80%

1 - 32 33 - 1023 1024 - 100000000

13%9%

78%

24%12%

64%

Number of Distinct Values

Inventory ManagementFinancial Accounting

% of

Colum

ns

20

Wide Tables Analysis of width of 144 most used* tables

* Largest in terms of cardinality

0

5

10

15

20

25

30

1-9

10-19

20-29

30-39

40-49

50-59

60-69

70-79 82

99

110-11

9 12

0-12

9 13

8 14

0-14

9 15

6 18

0-18

9 20

0-20

9 23

0 31

2 39

9

# Tables

# Columns

21

Learning Map

Founda(ons for a New Enterprise Applica(on

Development Era

Founda(ons of Database Storage

Techniques

The Future of Enterprise Compu(ng

Advanced Database Storage Tech-niques

In-Memory Database Operators

Changes in Hardware

Dr.-Ing. Jrgen Mller

Advances in Hardware

64 bit address space 4TB in current server boards

25GB/s data throughput, CPU - DRAM

Cost-performance ra8o rapidly declining

Mul8-Core Architecture 8 x (8-16) core CPU per blade

Parallel scaling across blades

One blade $50.000 = 1 Enterprise Class Server

A

24

Copy 50GB data via Inniband 10s

25

CPU Registers

Main Memory

Flash

Hard Disk

High

er

Perfo

rman

ceLower Price /

Higher Latency

CPU Caches

Memory Hierarchy

Latency Numbers L1 cache reference (cached data word) 0.5ns

Branch mispredict 5ns

L2 cache reference 7ns

Mutex lock/unlock 25ns

Main memory reference 100ns 0.1s

Send 2K bytes over 1 Gb/s network 20,000ns 20s

SSD random read 150,000ns 150s

Read 1 MB sequen8ally from memory 250,000ns 250s

Disk seek 10,000,000ns 10ms

Send packet CA to Netherlands to CA 150,000,000ns 150ms

26

Learning Map

Founda(ons for a New Enterprise Applica(on

Development Era

Founda(ons of Database Storage

Techniques

The Future of Enterprise Compu(ng

Advanced Database Storage Tech-niques

In-Memory Database Operators

A Blueprint of SanssouciDB

Dr.-Ing. Jrgen Mller

SanssouciDB: An In-Memory Database for Enterprise Applications

Main Memoryat Blade i

Log

SnapshotsPassive Data (History)

Non-VolatileMemory

RecoveryLoggingTime travel

Data aging

Query Execution Metadata TA Manager

Interface Services and Session Management

Distribution Layerat Blade i

Main Store DifferentialStore

Active Data

Me

rgeCo

lum

n

Co

lum

n

Co

mb

ined

Co

lum

n

Co

lum

n

Co

lum

n

Co

mb

ined

Co

lum

n

Indexes

Inverted

ObjectData Guide

In-Memory Database (IMDB)

Data resides permanently in main memory

Main Memory is the primary persistence

S8ll: logging to disk/ recovery from disk

Main memory access is the new bo\leneck

Cache-conscious algorithms/ data structures are crucial (locality is king)

Main Memory at Server i

Distribution Layer at Server i