Final presentasi gnome asia

An Overview, Maximize The Ability of

The GNU/Linux Operating System Using

"In Memory Computation"

for Academic, Business and Government

Anton Siswo Raharjo Ansori

A little of Myself

Member of KLAS (Komunitas Linux Arek Suroboyo)2008 - Now

Lecturer at Telkom UniversityMarch, 2015

Born : May, 11th 1987Live in Bandung, Come from Pasuruan

Mobile Number : +6282233932728Email : [email protected] [email protected]

mailto:[email protected]

mailto:[email protected]

Who had done it

Using application which whole processing is in memory

Using General OS

More and More Company

Image from http://www.welltech.in.ua/images/Linux_Family.jpg

Who had done it

Who had done it

Who had done it

Who had done it

Example

Let's talk about In Memory Computing

What is it?

Why we need it?

Where it can be applied?

Cost reduction and calculation

Computer Component

Network Topology with and Without HDD

How we build it

What next research

Technical Specs

Benchmarking

Live and Video Demo

What is In Memory Computing


Whole OS in Memory


All OS Processes


Critical Process Only


Become Storage as HDD


Whole OS in Memory

All OSProcesses

Critical Process Only

Become Storage as HDD

OR


Put simply, in-memory computing primarily relies on

keeping data in a server's RAM as a means of

processing at faster speeds. In-memory computing

especially applies to processing problems that

require extensive access to data–analytics, reporting

or data warehousing, and big data applications.

-- IBM --

Put it Simply

Put it Simply

Whole OS in Memory

Put it Simply

Whole OS in Memory

Put it Simply

Intention?

Whole OS in Memory

Put it Simply

Get the maximum performance with the minimum bottleneck

Intention?

Whole OS in Memory

CONS of In Memory Computing (1)

At this moment:

● Not yet implemented Data Sync between

“HOST” and “In Memory” computer

● Temporary file system

– data loss at electricity failure –

CONS of In Memory Computing (2)

● Limited Disk at max memory capacity in

16GB for single Notebook

● Mounting local disk manually

Why we need it

● Better, faster, decision making. This includes

the ability to reduce cost, identify competitive

opportunities, grow revenue, become more

efficient and reduce risk.

-- IBM --

● General and Specific Purposes

● System for Realtime processing

Why we need it

Green Technology

Easier Maintenance

Less Electricity, Less temperature

Minimum Drive and OS Error

Areas which need the most

Bottleneck of Storage, Memory and Processor in :

– Computing

– Intensive tasks

– Databases

Where it can be applied

Academic


Business


Government


Software Developer


Academic

Business

Government Software Developer

Business

More specific purposes

● In daily activity

● In research for HPC areas

● Lower cost and greener Labs

● In Public computer at campus


● For business which need realtime

processing

● For handling big data processing

● Using in memory computing to minimize

hardware maintenance and cost

reduction


● Faster compiling

● Testing system

● Developing

● Benchmarking

Common Computer Component

Common Computer Network Topology with HDD

Computer Computer Computer Computer


Computer

switch

Computer Computer Computer


switch

Server Server Server


Computer

switch

Server


Server Server

Proposed Computer Network

Topology For Non Storage Drive



switch

Computer

Server For Non-Storage Client


switch

Computer

switch



switch

Computer

switch



Server Web

Server Datacenter

switch

What next is Benchmarking

Phoronix Benchmarking

Disk Suite

Processor Suite

Memory Suite

Benchmarking

MemoryHDD

Benchmarking

SQLite Fail

HDD

PHORONIX-TEST-SUITE.COMMB/s, More Is Better

AIO-Stress v0.21Random Write

disk-51

700 1400 2100 2800 3500

SE +/- 84.953076.56

Phoronix Test Suite 5.6.0

1. (CC) gcc options: -pthread -laio

PHORONIX-TEST-SUITE.COMSeconds, Less Is Better

SQLite v3.7.312,500 INSERTs

disk-51

300 600 900 1200 1500

SE +/- 51.271420.52


1. (CC) gcc options: -O2 -ldl -lpthread


AIO-Stress v0.21Test: Random Write

aio-stress-mem

1200 2400 3600 4800 6000

SE +/- 50.865450.16


1. (CC) gcc options: -pthread -laio

Memory

PHORONIX-TEST-SUITE.COMFiles/s, More Is Better

FS-Mark v3.31000 Files, 1MB Size

disk-51

6 12 18 24 30

SE +/- 0.4323.23


1. (CC) gcc options: -static

Benchmarking

MemoryHDD

PHORONIX-TEST-SUITE.COMFiles/s, More Is Better

FS-Mark v3.3Test: 4000 Files, 32 Sub Dirs, 1MB Size

fs-mark-mem

1000 2000 3000 4000 5000

SE +/- 20.484871.33


1. (CC) gcc options: -static


Dbench v4.048 Clients

disk-51

20 40 60 80 100

SE +/- 1.1582.23


1. (CC) gcc options: -lpopt -O2



disk-51

20 40 60 80 100

SE +/- 1.2797.14




Dbench v4.0Client Count: 256

dbench-mem

1500 3000 4500 6000 7500

SE +/- 38.137176.82





disk-51

11 22 33 44 55

SE +/- 0.1348.62




IOzone v3.4058GB Read Performance

disk-51

2000 4000 6000 8000 10000

SE +/- 88.678136.12


1. (CC) gcc options: -O3


IOzone v3.4058GB Write Performance

disk-51

20 40 60 80 100

SE +/- 0.2982.65




IOzone v3.405Record Size: 1MB - File Size: 8GB - Disk Test: Read Performance

iozone-mem

2000 4000 6000 8000 10000

SE +/- 18.548598.22




IOzone v3.405Record Size: 1MB - File Size: 8GB - Disk Test: Write Performance

iozone-mem

1100 2200 3300 4400 5500

SE +/- 44.105299.84




Threaded I/O Tester v0.3.364MB Random Read - 32 Threads

disk-51

3000 6000 9000 12000 15000

SE +/- 86.2915038.89



Threaded I/O Tester v0.3.364MB Random Write - 32 Threads

disk-51

0.1215 0.243 0.3645 0.486 0.6075

SE +/- 0.000.54



Threaded I/O Tester v0.3.3Test: Random Read - Size Per Thread: 256MB - Thread Count: 32

thread-io-test-mem

3000 6000 9000 12000 15000

SE +/- 213.5913907.61



Compile Bench v0.6Test: Initial Create

disk-51

70 140 210 280 350

SE +/- 7.65303.04



Compile Bench v0.6Test: Read Compiled Tree

disk-51

200 400 600 800 1000

SE +/- 0.15955.16



Compile Bench v0.6Test: Compile

disk-51

130 260 390 520 650

SE +/- 1.05584.58


In Memory side - Failure


Unpacking The Linux Kernellinux-2.6.32.tar.bz2

disk-51

3 6 9 12 15

SE +/- 0.129.07


PHORONIX-TEST-SUITE.COMTPS, More Is Better

PostMark v1.51Disk Transaction Performance

disk-51

1200 2400 3600 4800 6000

SE +/- 38.335396




Unpacking The Linux Kernellinux-2.6.32.tar.bz2

unpack-kernel-mem

2 4 6 8 10

SE +/- 0.028.90


PHORONIX-TEST-SUITE.COMTPS, More Is Better

PostMark v1.51Disk Transaction Performance

postmark-mem

2000 4000 6000 8000 10000

SE +/- 95.678429



Benchmarking

HDD


Gzip Compression2GB File Compression

disk-51

3 6 9 12 15

SE +/- 0.6413.57


PHORONIX-TEST-SUITE.COMRequests Per Second, More Is Better

Apache Benchmark v2.4.7Static Web Page Serving

disk-51

8000 16000 24000 32000 40000

SE +/- 632.9239242.25


1. (CC) gcc options: -shared -fPIC -O2 -pthread


Gzip Compression2GB File Compression

gzip-mem

3 6 9 12 15

SE +/- 0.0112.95


PHORONIX-TEST-SUITE.COMRequests Per Second, More Is Better

Apache Benchmark v2.4.7Static Web Page Serving

apache-mem

9000 18000 27000 36000 45000

SE +/- 66.5641255.76


1. (CC) gcc options: -shared -fPIC -O2 -pthread

Benchmarking

Memory Suite in HDD

and

Memory Suite in Memory


RAMspeed SMP v3.5.0Integer Add

mem-suite-test

3000 6000 9000 12000 15000

SE +/- 0.0013818.12



RAMspeed SMP v3.5.0Integer Copy

mem-suite-test

3000 6000 9000 12000 15000

SE +/- 0.0014320.51



RAMspeed SMP v3.5.0Integer Add

mem-suite-mem

3000 6000 9000 12000 15000

SE +/- 0.0013813.87



RAMspeed SMP v3.5.0Integer Copy

mem-suite-mem

3000 6000 9000 12000 15000

SE +/- 0.0014221.83



RAMspeed SMP v3.5.0Integer Scale

mem-suite-test

3000 6000 9000 12000 15000

SE +/- 0.0014288.12



RAMspeed SMP v3.5.0Floating-Point Add

mem-suite-test

3000 6000 9000 12000 15000

SE +/- 0.0015690.93



RAMspeed SMP v3.5.0Integer Scale

mem-suite-mem

3000 6000 9000 12000 15000

SE +/- 0.0014202.47



RAMspeed SMP v3.5.0Floating-Point Add

mem-suite-mem

3000 6000 9000 12000 15000

SE +/- 0.0015538.87



Stream v2013-01-17Add

mem-suite-test

3000 6000 9000 12000 15000

SE +/- 7.7616334.69


1. (CC) gcc options: -O3 -march=native -fopenmp


Stream v2013-01-17Copy

mem-suite-test

3000 6000 9000 12000 15000

SE +/- 9.1014579.18




Stream v2013-01-17Add

mem-suite-mem

3000 6000 9000 12000 15000

SE +/- 17.7716215.08




Stream v2013-01-17Copy

mem-suite-mem

3000 6000 9000 12000 15000

SE +/- 11.7214462.49




Stream v2013-01-17Scale

mem-suite-test

3000 6000 9000 12000 15000

SE +/- 6.2014485.59




CacheBenchRead Cache

mem-suite-test

600 1200 1800 2400 3000

SE +/- 3.992785.24


1. (CC) gcc options: -lrt


Stream v2013-01-17Scale

mem-suite-mem

3000 6000 9000 12000 15000

SE +/- 13.7714357.91




CacheBenchRead Cache

mem-suite-mem

600 1200 1800 2400 3000

SE +/- 0.142791.91




CacheBenchWrite Cache

mem-suite-test

3000 6000 9000 12000 15000

SE +/- 11.0814896.86




CacheBenchWrite Cache

mem-suite-mem

3000 6000 9000 12000 15000

SE +/- 7.2014910.62



Benchmarking

MemoryHDDmem-suite

RAMspeed SMP

RAMspeed SMP

RAMspeed SMP

RAMspeed SMP

Stream

Stream

Stream

CacheBench

CacheBench

mem

-su

ite-te

st

13818.12

14320.51

14288.12

15690.93

16334.69

14579.18

14485.59

2785.24

14896.86

PHORONIX-TEST-SUITE.COM

mem-suite-mem

RAMspeed SMP

RAMspeed SMP

RAMspeed SMP

RAMspeed SMP

Stream

Stream

Stream

CacheBench

CacheBench

mem

-su

ite-m

em

13813.87

14221.83

14202.47

15538.87

16215.08

14462.49

14357.91

2791.91

14910.62

PHORONIX-TEST-SUITE.COM

Processor Benchmarking

Memory

HDD


Smallpt v1.0Global Illumination Renderer; 100 Samples

smalpt

20 40 60 80 100

SE +/- 1.15104


1. (CXX) g++ options: -fopenmp


Smallpt v1.0Global Illumination Renderer; 100 Samples

smalpt-mem

20 40 60 80 100

SE +/- 0.88104


1. (CXX) g++ options: -fopenmp

How we build it

Linux In Memory Computing

How we build it

LO

CK

Install

How we build it

LO

CK

Load Linux to Memory at Start Up

How we build it

When Finish, Your Linux at Your RAM

How we build it

LO

CK

Let's demo ItBuilding in GNU/Linux OS

Let's demo It

Transfer session at boot up cost up to 10 minutes for 5.3 GB

Let's demo It

Proof of “In Memory Computing”

We are Searching for Partnership

Research

Project

and

Collaboration

Tribute to:

● KLAS (Komunitas Linux Arek Suroboyo) –

klas.or.id

● Tel-U (Telkom University) –

telkomuniversity.ac.id

● Free and Open Source Licence image

● Gnome Asia Summit 2015

● Everyone who cannot be written their name in

here

Sayonara....

Thank You....

Terima Kasih....

Question Answer

and

Improvement Sections

Final presentasi gnome asia

Presentations & Public Speaking