Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC...

15
Globus Toolkit Massimo Sgaravatto INFN Padova

Transcript of Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC...

Page 1: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Globus Toolkit

Massimo SgaravattoINFN Padova

Page 2: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Introduction Grid Services: LHC regional centres

need distributed computing Analyze toolkits that integrate

distributed computational resources: Globus Legion Globe ….

Page 3: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Globus Basic research on GRID problems Globus Toolkit

“Bag of services” for developing GRID tools and applications

Available for most Unix platforms Limited functionalities for Win32

Page 4: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Globus ArchitectureApplications

Core ServicesMetacomputing

Directory Service

GRAMGlobus

Security Interface

Heartbeat Monitor

Nexus

Gloperf

Local ServicesLSF

Condor MPI

NQEEasy

TCP

SolarisIrixAIX

UDP

High-level Services and ToolsDUROC globusrunMPI Nimrod/GMPI-IO CC++

GlobusView Testbed Status

GASS

Page 5: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Core Services Security (GSI) Information Service (MDS) Resource Management (GRAM) Remote Data Access (GASS) Communication (Nexus) Network Performance Monitor (GLOPERF) Fault detection (HBM) Executable Management (GEM) (???)

Page 6: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Globus Security Infrastructure Main focus: authentication

single authentication not necessary to manage authentication

on each GRID node Mapping between GRID user-id and

local user-id Digital signature mechanisms

Page 7: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Metacomputing Directory Service

Static and dynamic information about GRID components (host architecture, memory, load, network bandwidth, latency. ...)

Possibility to define new attributes Information updated by:

Globus services (i.e. GRAM Reporter) Users/Applications Other “providers” (i.e. SNMP)

Tools and API for discovering, publishing, and accessing information

Standard protocol and interface: LDAP Single MDS server (located in USA)

Page 8: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Resource Management Management of computational resources GRAM (Globus Resource Allocation Manager)

responsible for a set of “local” resources Single interface to different resource managers

(Condor, LSF, NQE, fork, ...) Resource Specification Language (RSL) for

defining: resource requirements (machine type, memory, ...) job definition (executable, parameters, ...) I.e.: & (ResourceManagementContact = “cond.pd.infn.it:/C=US/.../CN=cond.pd.infn.it- condor”) (count=100) (executable=myapp)

Page 9: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

GRAM

CONDOR

Globus Resource Manager Architecture

Application

BrokerMDS

GRAM

LSF

GRAM

fork

Resource andJob specification

Resourcediscovery

Create jobs(Ground RSL expressions)

Site1Site2 Site3

(High level RSL expression)

RSL specialization

Page 10: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Remote Data Access GASS (Global Access to Secondary Storage) Supported servers: GASS, HTTP, FTP Uniform naming using URL:

I.e.: x-gass://host1.pd.infn.it:1234/filename Possibility to access remote data sets (distributed

file system) Necessary to modify source code:

open() globus_gass_open() close() globus_gass_close()

Caching Possibility to move files from/to remote data

storage systems

Page 11: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Other core services Nexus

Unicast and multicast low level communication library

Supports different programming models (message passing, RPC, etc…)

HBM (Heartbeat Monitor) Monitors the status of registered processes

GLOPERF (Globus Network Performance Measurement Tool)

Measures throughput and latency between node pairs

Page 12: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Evaluation of the Globus Toolkit

5 sites Testbed (Bologna, CNAF, LNL, Padova, Roma1)

Use case: HLT CMS studies MC Prod. Complete HLT chain

Services Resource management

fork() Interface to different local resource managers (Condor, LSF)

Resources chosen by hand Broker to implement a Global resource manager

Data Mover to stage executable and input files to retrieve output files

Page 13: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Use Case: CMS HLT studies

Page 14: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Status Globus installed on 5 Linux PCs in 3

sites GSI

Seems working MDS

Lots of problems, maybe temporary? GRAM, GASS, Gloperf, HBM

Work in progress

Page 15: Globus Toolkit Massimo Sgaravatto INFN Padova. Massimo Sgaravatto Introduction Grid Services: LHC regional centres need distributed computing Analyze.

Massimo Sgaravatto

Other problems Documentation not well organized Collaboration with HPC Lab,

University of Lecce is very useful !