Operational Architecture of PLGrid project

1
Operational Architectureof PL-Grid project

• M.Radecki, T.Szepieniec, M.Krakowian, M.Tomanek
  and W.Ziajka
• ACC CYFRONET AGH

Cracow Grid Workshop Cracow, 12.10.2009
2
Outline

• Customers in Grid and issues to be solved
• VO manager
• Resource Provider
• VO user
• Site Administrator
• PL-Grid approach to resource sharing
• Bazaar tool
• Other operational consequences
• Information monitoring system improvements
• Solving operational issues in PL-Grid

3
Grid Customers and Issues

• Virtual Organization Manager
• how could I obtain (more) resources for my VO?
• does QoS for provided resources satisfiy me?
• VO User
• I want to use the right resources i.e. these
  which are well configured and supported
• Resource Provider
• Are all VOs satisfied with resources they get?
• Site Administrator
• ooops, something failed in grid middleware at my
  site, how to fix it quickly?

4
Linking Customer with Provider

• Well-defined procedure for VO Managers and
  Resource Providers for making an agreement on
  sharing resources is indispensable

5
Resource-related SLA and Bazaar tool

• PL-Grid will use Bazaar tool for implementing
  process of resource negotiation between VO
  manager and Resouce Provider
• Example contract parameters
• time boundaries of the contract
• number of CPUs, disk space with the Quality of
  Service (guaranteed, best effort etc.)
• availablity/reliability of resources
• average acknowledge/response time to trouble
  tickets

6
Profits from having resource-related SLAs

• Communication channel between parties established
• Agreement can be monitored
• need accounting data
• monitoring availability/reliability of services
• trouble ticket ack/response time
• First step towards business model in grid
• Impact on operations
• verification (certification) of resources for
  particular contract
• only certified resources accessible for VO users

7
Infrastructure Monitoring

• Typically one dedicated VO for monitoring all
  resources in Grid (e.g. ops VO in EGEE)
• requirement on sites to support this VO for
  monitoring purposes
• configured as high-priority VO
• not always reflects the status of the site
  according to other VOs
• PL-Grid uses regular VOs for monitoring
• special role configured within VO
• high priority for jobs executed with this role
• requires subscription of a technican's
  certificate as VO member
• reflects the VO status at a site
• site's service availability/reliability measured
  within real VOs

8
Grid Information System Improvements

• Use VO-level Information Systems instead of the
  global instance
• VO-scope makes sense for user
• better scalability no longer global grid service
• easier to manage
• can be handled by the VO as other VO-services VO
  Membership Service, File Catalogue, Resource
  Broker, etc.
• big VOs may take an effort to establish a
  high-availability cluster for information
  service, smaller ones will not require that
• reduces the network traffic by localizing it
• Include information only about sites which have
  an active contract with the VO and resources were
  verified (certified)
• Require a separate instance of Information System
  including all sites for testing/certification
  purposes

9
Solving Grid Operations Problems

• Site Administrator's perspective
• many sources of data wiki pages, GGUS knowledge
  base
• many of them outdated, not providing working
  recipe
• Customers (VOs) are pressing on
  availability/reliability of resources
• need for quick problem solving
• need for interactive support e-mail not always
  efficient
• PL-Grid support structures
• Actors
• Site Administrators
• 1st line support Team
• Regional Operator on Duty (aka ROD)

10
Use case Operations Problem Handling
Monday, 7 P.M.
Tuesday, 8 A.M.
Tuesday, 9 A.M.
Tuesday, 7 P.M. 24h passed
Wednesday, 8 A.M.
Site
11
Summary

• Identified need for new operational tool for
  Resource Allocation
• fill a gap between VO Managers and Resource
  Providers
• Improvements to Information System related to RA
  and VO-scope
• to provide VO User with list of reliable,
  well-configured and supported resources
• Infrastructure monitoring should be realized
  within the real, existing VOs
• Procedures for fixing problems can be more
  efficient with
• knowledge base to find if somebody else
  encountered the problem beforehttps//weblog.plgr
  id.pl/baza-wiedzy/
• 1st line support team to get interactive contact
  with the expert
• Polish NGI got assigned share in EGI global tasks
  related to
• Coordination of resource allocation (O-E-10) -
  Poland
• Grid Operation and oversight (O-E-5) -
  Netherlands and Poland

12
PL-Grid news (tentative) user registration open!

• http//www.plgrid.pl/oferta/infrastruktura/wyprob
  uj_grid