Polish Infrastructure for Supporting Computational Science in the European Research Space EUROPEAN UNION Services and Operations in Polish NGI M. Radecki,

Polish Infrastructurefor Supporting Computational Science

in the European Research Space

EUROPEAN UNION

Services and Operationsin Polish NGI

M. RadeckiM. Radecki, T. Szymocha,, T. Szymocha,T. Szepieniec, M. PawlikT. Szepieniec, M. Pawlik

ACC Cyfronet AGHACC Cyfronet AGH

Cracow Grid Workshop9 Nov 2011

2

OutlineOutline

Users and Resources

Services for Users

Guarantees - Service Level Agreement

Supporting SLAs in Operations

3

Users & ResourcesUsers & Resources

Polish researchers PhD owners – base user group,

must confirm affiliation with Polish research institution

MSc & PhD students have to find their supervisor who

confirms collaboration on research subject

International collaborations need a collaboration with Polish

researchers (similar to supervisor) international VOs here

Computing: ~23k cores

Storage: ~2PB

Each site has autonomy in

resource allocation to users

All site resources accessible for all middlewares

gLite

UNICORE

QosCosGrid

Local batch system

Numbers (as of 1st Nov):

782 users, 4.4 Mhours of computing in October47% LHC VOs, 53% PL-Grid users (in October 2011 only)

4

Model of Delivering ServicesModel of Delivering Services

Service LevelAgreement

Service LevelAgreementAccess ServiceAccess Service Metrics

Access

Use C

onditions

+

Guaran

tees

Access Service

Resources

5

„„Access Services” – access to resourcesAccess Services” – access to resources

Global access gLite UNICORE QosCosGrid

Local access Batch system MySQL GPGPU vSMP User Interface machine – at each site, clients for all middlewares

6

Stages of service use – user viewStages of service use – user view

1. Become PL-Grid user

1. Get credentials - easy, on-line

2. Request access

1. User applies for activating access to the service

2. Application is answered by the service administrator

3. Service admin can manage the access afterwards

3. Establish SLA

4. Use

1. User should be able to observe service status and their current usage

2. Overuse is blocked (fair play)

5. Account

1. User accounts for consumed resources

7

Lesson learnt: pass auth* info to serviceLesson learnt: pass auth* info to service

Pass access info from central point to the service, securely, reliably what if the central database does not work what if it desynchronizes

8

SLAs for users = computing grantSLAs for users = computing grant

Why? – to build relation between Provider and User Provider need to know users expectation – necessary for resource use

planning User need to share their plan to get any guarantees on what they want – more

guarantees more user satisfaction What is needed?

Grant submission

• Researcher puts objectives, expected results, resources, additional services

• Review - grant evaluation a commitee who gives recommendation support decision at sites

Grant negotiation Resource Allocation – setup of resources according to SLA Resource Use Monitoring

• observe use, make sure agreed tresholds are met, block if overuse Accounting

9

Grants as implemented in PL-GridGrants as implemented in PL-Grid

Two types of grants depending on size of requested resources Personal grant – testing, trying – 6 months, 1000h walltime, 40GB Regular grant – intensive computing here

Tools: Portal + Bazaar Grant belongs to User Team

self-organizing all members can use grant

Resource Allocation – Bazaar Site Admin Toolkit

takes grant details from Bazaar and generates site scheduler config, run daily

site administrator retain full control over their services

Accounting – User Portal each 6 months a short report final report after grant finishes

10

Grant MonitoringGrant Monitoring

Computing Resource Use monitoring system resources consumed within given grant notion of middlewares – match jobs as being executed within specific mware

Matching Job and Grant ID – user must declare during job submission – e.g in JDL „default” grant declared in Portal – all jobs account on this if not specified else

Support for grants in gLite, UNICORE, QosCosGrid UNICORE – under development (XSEDE requirement) gLite – requirement submitted: EGI RT #2983

• temporal workaround:use VO_TAGS

QCG – under development Show consumed resources

per grant in User Portal walltime, number of jobs,

site – daily stats.

11

Important „elements” of IT Service ManagementImportant „elements” of IT Service Management

Policies – general guidelines and objectives defined by PL-Grid Strategy Team

Processes – sets of interrelated activities that converts input into output

Procedures – specified ways to perform activities

Plans People, teams Tools Other resources: budget, technologies

© T.Schaaf „Tutorial: Towards better managed Grids. IT Service Management best practices based on ITIL”

12

Processes in PL-Grid OperationsProcesses in PL-Grid Operations

User and Group Management User/Group Registration, de-

registration Access Management

granting access to service blocking access to service

Service Level Management Handling incoming SLA app. SLA Monitoring SLA Accounting

Service Availability Management Service Availability Monitoring Technical Support for

Administrators Knowledge Base Maintenance

User Support

Handling tickets

Internal knowledge DB maintenance

User FAQ maintenance

New application deployment

User documentation maint.

Configuration Management

Sites

Services

Downtimes

Change Management

Known type change mgmt - follow procedure

New type change mgmt – evaluation, discussion, decision, write new procedure

13

SummarySummary

PL-Grid Infrastructure has rich service offer to users

Users can get some guarantees together with access to resources

Idea of „computing grants” is essential for making step forward in PL-Grid service level

Well organized Operations is a key factor for successful SLA adoption

14

Thanks for your attention!Thanks for your attention!