D-Grid Infrastructure

D-Grid Infrastructure

Stefan FreitagDortmund University of Technology

6th International Workshop on Grid Computing for Complex Problems

Bratislava, November 2010

Outline

● Projects & Infrastructure● Lessons learned & Future Challenges

2004 2007 20102009 20112008 2012

Gap project

20062005

Integration project

Service layer

Academic projects

Public/ private projects

D-Grid GmbH

Working Groups

Pre-project Phase 1

Phase 2

Phase 3

D-Grid Projects

Gap projects

Service layerprojects

Commercial communityprojects

Academic communityprojects

Integration project

Project structure of D-Grid

Representation (?)

Academic service and resource providers

Academic research communities

D – Grid GmbH

Industrial service and resource providers

Organizational issues

Legal issuesInternational

issues

Public Relation

GAUSS-Allianz WissGrid

Representation structure in D-Grid

Academic projects

Astronomy

Climate Research

High Energy Physics

Medical Science

Humanities

Aircraft Industry

Financial Business

Automobile Industry

Building & Construction Industry

Geospatial Information Systems

Medical Services

Meteorology for regenerative energy industry

Commercial Projects

Commercial Projects

Logistics

Media Technology

Chemical Industry

Numerical Simulation

Medical Engineering

Plasma Technology

Dynamic Service Value Networks

Service Projects

Backup & Recovery

Sustainable Grid Infrastructure

SLA4D-Grid

Networks of Knowledge in the Grid

Gap & Service Projects

VO management

D-MON

DGSI

GapSLC

GIDS

D-Grid Infrastructure

Situation in 2005

High energy physicists● advanced knowledge (CERN experiments)● gLite middleware

Other communities● not really any prior knowledge

How to make all communities happy?● Let them choose!● Community requirements still unclear● Offer a set of commonly used middlewares

Leere Fusszeile

Supported Middlewares

Compute middleware● gLite● UNICORE● Globus Toolkit

Storage middleware● dCache● OGSA-DAI

Leere Fusszeile

Software Stack @ D-Grid Resource

Leere Fusszeile

D-Grid Reference System

● Template for a D-Grid compute and storage resource

● Targeted at small and medium resource centers

● Technical documentation & software packages available to system administrators

● http://dgiref.d-grid.de/wiki/Introduction

● Validation of packages and documentation by installation “from scratch”

● Release Cycle: six months

● Reference system is only a recommendation

● Support available

D-Grid Resources

● > 30.000 CPU cores

● > 5 PByte of storage

● In 2009 >= 140 Mio. CPUh consumed by users

● Accounting ● DGAS/ manually

Usage of D-Grid Resources

Towards Sustainability

Technology is not really a problem!

Focus on existing concepts

● Support for different Grid middlewares

● Virtualization

Bridging the gap to services and applications

● Workflow engines

● Commercial software

Addressing new requirements

● Dynamic virtual organizations

Cooperation between different projects

● Scheduling

Heterogeneous Provider Community

Some large hardware providers play a vital role in D-Grid

● Development of a business model for basic services

Some providers used D-Grid to acquire new hardware

● Investments to support a transitional phase to Grid systems

No known concepts to provide IaaS type of services

● Obstacle for commercial users

● Competition to cloud computing?

There are not enough users!

Prominent figures of the various disciplines have not yet joined● Imbalance between computer science and application science

Most users are not willing to provide the required additional effort. ● Potential lack of focus in the application area

Industrial users are pushed off by bureaucratic obstacles● Use of D-Grid hardware for commercial purpose

Grid computing is a solution, isn’t it?

Funding agencies cannot be wrong!

● More than 100 Mio € in D-Grid

Industry cannot be wrong!

● Enterprise Grids

● Cloud Computing

Science cannot be wrong!

● LHC production Grid

Due to the weather conditions... Cloud Computing!

Photo by: Paolo Malgari

Cloud Computing in D-Grid ?

Situation in 2009● Utilization of compute resources ≥ 140 Mio CPUh

● only around 50% of the available CPUh

½ of D-Grid compute resources idling● Turn them off → GreenIT

● Not really acceptable for a sustainable operation

→ Introduce compute cloud interface as new pillar

Why ?

● Grid is too complex and hence not attractive to SMEs

Attractivity of Cloud Interface I

Cloud● Register once via web. Specify name, credit card

information,... (few minutes)

Grid● Application for X.509 certificate (1-2 days)

Contact Certification authority

Contact Registration authority (f2f required)● Application for VO membership (hours - days)● Updating of the authorization information on the Grid

resources (hours - days)

Attractivity of Cloud Interface II

Cloud Computing in D-Grid

● Extension of the software stack

● What to add?● Platform

virtualization● Compute cloud

middleware

D-Grid Cloud Prototype

Located at Dortmund University of Technology

Services● Compute middlewares

– gLite 3.2

– UNICORE 6

– Globus Toolkit 4.0.8

● Storage middlewares

● dCache 1.9.5

● OGSA-DAI WSRF 2.2

Hardware● 2048 CPU cores, 108 TByte storage

All services deployed in Xen virtual machines

OpenNebula

Statement on homepage● Standard-based open-source toolkit to build private, public and

hybrid clouds. [...] provides flexible architecture, interfaces and components that fit into any existing data center.

Supports Xen, KVM and VMware

Interfaces with Amazon EC2 and ElasticHosts

Other users● CERN (cloud infrastructure consisting of 3,200 cores)

● NIKHEF, Netherlands

Prototyp Setup

Open Issues

Technical level● User management● Authorization● Accounting and Billing● Monitoring● Information system

→ Interfaces to all D-Grid core services

User Management

D-Grid● VO central user management

● Resources query VOMRS servers and generate a mapping

● VO member → local user ID

● Attribute-based authorization

Cloud (OpenNebula)● Local user management,

● No Attribute-based authorization

Can the concept of VOs and Cloud middleware go together?

Lessons learned

Reference System Acceptance

Only a few sites ● Follow the recommended setup● Use the technical documentation

Why should a resource install all middlewares?● System administrators are not qualified to support all

middlewares ● Lack of interest at resource provider level

Migration/ Upgrade path unspecified

Legal Issues – Dual Use

Dual-use technology: a technology which can be used for both peaceful and military aims.

Most industrial countries have export controls on certain types of designated dual-use technologies […] These controls restrict the export of certain commodities and technologies without the permission of the government.

(Wikipedia)

● Is Grid a dual-use technology?

● Export embargo for dual-use technologies to Iran

● Technical solution for a legal requirement

Legal Issues – Liability

With the development of business models, the liability of single VO members and of the whole VO is of interest

What is the legal status of a virtual organization?● Company constituted under Civil law

→ Joint and several liability● In joint and several agreements, the liability for default is

enforceable against all of the signatories as a group[...], or against any one of them as an individual at the choice of the enforcing party (www.businessdictionary.com)

Can business models and virtual organizations co-exist?

Avoid Parallel Universes

Before D-Grid German HEPs used WLCG resources

In 2010 German HEPs still use WLCG resources

● and not a single one uses D-Grid resources

Why?

● Accessibility to compute resources, data and experiment software is limited by the boundaries of a virtual organization

● In WLCG VOs for ALICE, ATLAS, CMS and LHCb were already existing

● In D-Grid a HEP-CG VO was established with the aim to absorb all HEPs independent from their origin

● Even if HEPs joined the HEP-CG VO, they were not able to access the experiment software

Future Challenges

Future Challenge: D-Grid/ NGI-DE

EGI was still EGEE as D-Grid started

With the end of EGEE-III, NGI-DE was established

Since then D-Grid and NGI-DE co-exist in Germany

Merge to only one initiative

● Avoid competition

● Combine endeavors for creating a sustainable infrastructure

● Removal of duplicity of service (GRRS/ GOC DB)

● Integration of D-Grid services (D-Mon)

● Operations concept (1:1 mapping, pool accounts)

New Service: D-Mon

Summary

Projects in D-Grid

D-Grid Infrastructure● Software stack, resources● Prototypic integration of a Cloud middleware

Lessons learned

Legal issues & Future challenges

Thank you for your attention!

If you found this talk interesting, slides are available at

http://www.slideshare.net/stefanfreitag