gLite The Next Generation Grid Middleware Lukasz Skital

1
gLiteThe Next Generation (?) Grid Middleware
Lukasz Skital
3 listopad 2005
EGEE is a project funded by the European Union
under contract IST-2003-508833
2
Outline

• Introduction
• Middleware
• Architecture
• gLite Services
• Information and Monitoring System (R-GMA)
• Service Discovery (SD)
• Virtual Organisations Management System (VOMS)
• Workload Manager, Computing Element and Worker
  Nodes
• Logging and Bookkeeping
• Data Catalogs, File Transfer etc.
• gLite User Guide
• Job submission
• Job information
• Data management
• gLite in Cyfronet
• EGEE preproduction system

3
IntroductionMiddleware

• Middleware makes Grid from distributed resources
• Layer between user (application) and resources
• gLite EGEEs Next Generation Grid Middleware
• Provenance LCG (VDT,EDG), AlieEn,

4
Architecture
5
gLite Services
6
Information and Monitoring System(R-GMA)

• Implements GGF GMA standard
• Virtual database (SQL-like query language)
• Flexible, but slow

Application code
Consumer
Consumer servlet
Consumer API
lookup
Registry API
Registry servlet
Registry
Registry API
Schema API
Producer
Sensor code
Producer servlet
subscribe
Schema servlet
Producer API
7
Information and Monitoring System(R-GMA)
Publish Tuples
Producer Service
Producer application
API
Register
SQL INSERT
Registry Service
Tuples
Query
SQL SELECT
Locate
Schema API
Send Query
Consumer Service
Consumer application
API
Schema Service
Receive Tuples
SQL CREATE TABLE
8
Service Discovery (SD)

• gLite services database
• Can by used by end user and other services
• Compatible with
• R-GMA
• BDII
• File

9
Virtual Organizations Management System (VOMS)

• Derived from DataTag and EDG
• Used for Virtual Organizations management
• Virtual Organization
• Users
• Resources
• Resources can be assigned to multiple VOs
• VOMS certificate appended to user proxy
  certificate
• Understood by Workload Management and Data
  Management
• Allows multiple VO membership

10
Job Management

• WMS Workload Management System, resource broker
• CE Computing Element, interface to local
  computing resources
• WN Worker nodes
• Logging Bookkeeping job execution history
• Packet Manager dynamic installation of
  application software

11
Workload Management System (WMS)

• Resource broker
• Push mode
• WMS has a list of CEs
• WMS submits job to selected CE
• Necessary interaction with R-GMA (slow)
• Pull Mode
• CE has a list of WMS
• When CE has available resources, contacts WMS to
  get a job.
• Logging Bookkeeping

12
Workload Management System (WMS)
13
Computing Element (CE)

• Interface to local computational resources (WNs)
• Almost compatible with LCG one
• Two modes pull and push
• Condor client (WMSlt-gtCE)
• Local Resource Managment System (CElt-gtWN)
• Pbs (default Torque)
• Lsf
• Condor

14
Data Management

• Three main services group
• Catalog Services
• Core service
• Independent data catalogs for VOs
• Storage Services
• Local Storage Resource Manager (eg. Castor,
  dCache)
• gLite IO-server
• File Transfer Services
• File transfer
• File placement

15
Data ManagementCatalog

• File Catalog
• Filesystem-like view on logical file names
• Keeps track of sites where data is stored
• Conflict resolution

16
Data ManagementStroage Element

• SRM interface
• Management and control
• SRM (with possible evolution)
• Posix-like File I/O
• File Access
• Open, read, write
• Not real posix (like rfio)

17
gLite User Guide
18
gLite User Guide How to submit a job?

• Prepare a job (JDL file)

19
gLite User GuideHow to submit a job?

• Generate proxy certificate

20
gLite User GuideHow to submit a job?

• Submit the job (prepared JDL file)

21
gLite User GuideHow to submit a job?

• Check the job status

22
gLite User GuideHow to submit a job?

• Check the job status history (1/2)

23
gLite User GuideHow to submit a job?

• Check the job status history (2/2)

24
gLite User GuideHow to submit a job?

• Check the job status (2)

25
gLite User GuideHow to submit a job?

• Get a job output

26
DAG Jobs Directed Acyclic Graph

• Jobs with dependencies
• Allows to define data flow using Storage Element

• Assures job synchronization according to DAG
• DAG definded in JDL

27
gLite User GuideData Management

• Available glite-catalog operations

28
gLite User GuideData Management

• List directory content

29
gLite User GuideData Management

• List directory content

30
gLite User GuideData Management

• Stat a file

31
gLite User GuideData Management

• glite-get glite-put

32
gLite in Cyfronet
33
gLitein ACC Cyfronet AGH

• Part of EGEE pre-production service
• Resources
• ONE node zeus32,
• 2 CPU 2.4GHz, 2GB RAM, 120GB HDD
• Requirements
• 5 nodes
• Solution Virtualisation
• The Xen virtual machine monitor
• 9 virtual machines running on one real node
• Performance??? Good enough for pre-production

34
Cyfronet in PPS wiki
35
gLite in ACC Cyfronet AGH

• Nodes and services
• zeus32
• Xen Domain-0 direct access to hardware
• nfsd, dhcpd
• R-GMA Server
• zeus73 Storage element DPM
• zeus74 IO server
• zeus75 WMS (pull mode)
• zeus76 Computing elements/Worker node
• zeus77-79 Worker nodes
• zeus80 User Interface

36
References

• http//glite.web.cern.ch/glite/
• https//wiki.egee.cesga.es1234/cgi-bin/moin.cgi/C
  yfronetNodeList
• Comparison of LCG-2 and gLite, E.
  Slabospitskaya, IHEP
• gLite for ATLAS Production, Simone Campana,
  CERN/INFN