Title: Globus Overview
1Globus Overview
- CS-780-3 Lecture Notes
-
- In courtesy of Andy Kowalski
2Overview
- What Defines a Grid
- Globus
- GT1, GT2 (pre OGSA)
- GT3 (post OGSA)
- Issues
3What Defines a Grid?
- We will probably see the spread of computer
utilities, which, like present electric and
telephone utilities, will service individual
homes and offices across the country - -Len Kleinrock. 1969 1
4What Defines a Grid?
- A computational grid is a hardware and software
infrastructure that provides dependable,
consistent, pervasive, and inexpensive access to
high-end computational capabilities. - - Ian Foster and Carl Kesselman, 1998 2
5Ian Fosters Three Point Checklist (2002)
- A Grid is a system that
- coordinates resources that are not subject to
centralized control 2 - using standard, open, general-purpose protocols
and interfaces 2 - to deliver nontrivial quantities of service 2
6Globus Project
- Multi-institutional research effort on the
technology required to build a grid or grids - Mathematics and Computer Science Division at
Argonne National Laboratory - Information Sciences Institution at the
University of Southern California - University of Chicago
- University of Edinburgh
- Center for Parallel Computers
- Principle Investigators
- Ian Foster - Argonne National Laboratory and
University of Chicago - Carl Kesselman - University of Southern California
7Globus Project
- Seek to understand application requirements for a
usable grid and - Conduct research and development to create the
technologies required to build a grid - Globus toolkit
8Globus Toolkit
- Bag of services
- Developers can use these to build grid
applications - Makes use of existing standards when possible
- Interfaces and implementations
- Distinguishes between local and global services
- Defines interfaces to manage heterogeneity
- Does not hide heterogeneity
- Information service to deal with constant changes
- Utilization, failures, new resources, etc.
9Hourglass Principle
- A layered architecture with and hourglass shape
- Grid services should provide a simple
well-defined interface between applications and
local services
10GRAM
- Globus Resource Allocation Manager
- Provides a single interface for requesting and
using remote system resources for executing jobs - Each GRAM is responsible for the resource
management systems local to it - LSF (Load Sharing Facility), Condor, PBS
- Provides a standard interface to these different
systems - Applications can access these different systems
using a standard API - Within GRAM resource requests are described by
the Resource Specification Language (RSL)
11GRAM
Broker
RSL
Application
Information Service
Queries Info
Ground RSL
Co-allocator
Single Ground RSL
- Ground RSL is an expanded description of the
applications request. It may represent more
than on job. - Single Ground RSL is a description of a singe
job created from the Ground RSL. There may be
more than one created from the original Ground
RSL.
GRAM
GRAM
LSF
PBS
12RSL
- Resource Specification Language
- Provides a common language to describe resources
- Uses attribute, value pairs
- A simple example
- (executable a.out ( lt-- that is an unquoted
literal )) - (directory /home/nobody )
- (arguments arg1 "arg 2")
- (count 1)
13Nexus
- Low-level communications API that supports other
communication libraries - Can select alternative low-level protocols based
on network topology - Message passing, IP, shared memory
- Can introduce encryption based on source and
destination of a message - Intended for use by compiler writers and library
developers - CC, HPC, and MPICH-G
14MDS
- Metacomputing Directory Service (GT1)
- Monitoring and Discovery Service (GT23)
- Provides information about the systems in the
grid - CPU type, memory, OS, network, etc.
- Queue information, utilization, etc
- Basically a local LDAP server
- MDS ties all the LDAP servers together so
everyone has access to the same information - Has an API for discovering, publishing, and
accessing information
15MDS Architechture
- Grid Index Information Service (GIIS)
- Provides an aggregate directory of lower level
data - Grid Resource Information Service (GRIS)
- Runs on a resource and acts like a content
gateway to the resource - Information Provider (IP)
- Interface to data collection services
16GSI
- Globus Security Infrastructure
- Deals only with authentication
- Users and services
- Uses x.509 certificates
- Also used username and password (rlogin)
- Uses Secure Socket Layer (SSL) for the connection
and authentication - No longer true with GT3
- All Globus services use GSI to authenticate users
17GSI
- Use proxy certificates for non-interactive work
- Jobs executing on remote systems
- Proxy certificates have a limited life time
18CAS
- Community Authorization Service
- In the CAS model, resource providers grant access
to blocks of resources to a community as a whole,
and the community uses a CAS server to perform
fine-grained access control on those resources. - Resource providers grant course-grained access to
communities. - Communities run CAS servers, which keep track of
fine-grained access control information and grant
restricted proxies to community members. - The result is that a CAS user gets the
intersection of the rights granted by resource
provider to the community and the rights granted
by the community to that user.
Slide Taken From 6
19A Typical CAS Request
- CAS request, authenticated with
CAS Server
CAS-maintained community policy database
User credential
What rights does the community grant to this
user?
- CAS reply, including
- restricted proxy cred
Community subject name
Policy restrictions
User
Resource Server
3. Resource request, authenticated with CAS proxy
Is this request authorized for the community?
Local policy information
Community subject name
Policy restrictions
4. Resource reply
Do the proxy restrictions authorize this request?
Slide Taken From 6
20HBM
- Heartbeat Monitor
- API to register with the service and collect data
- Service expects updates from clients that have
registered - If no update is received, attempts to determine
if the service itself is down or if there are
other factors
21GASS
- Global Access to Secondary Storage
- Allows programs to use the standard C I/O library
to read and write files on remote computers - Reads
- File is copied locally and then read
- Writes
- File is written locally and then copied to their
destination - Appends
- Communicated to the remote file as soon as it is
written
22GEM
- Globus Executable Management
- Support identification, location, and creation of
executables in heterogeneous environments - Pre-stages executables to multiple remote sites
- Removes them after execution
23GridFTP
- Secure data transfer protocol
- GSI authentication
- Optimized for high-bandwidh wide-area networks
- Parallel streams - multiple data channels
- Partial file transfer
- Third-party transfers
- Server to server
24Globus Toolkit 3
- Based on the Open Grid Services Architecture
(OGSA) - Attempts to leverage industry standards
- Makes use of web services
- Preserves backward compatibility with GT2
- Future releases will begin eliminating
compatibility - Client interfaces to services are now WSDL/OGSI
based
25Issues
- Protocols and interfaces
- Where the GT1 and GT2 implementation of a grid
service was inadequate for certain grids, others
would developed competing services using
different interfaces and protocols - Defeated the goal of compatibility
- OGSA/OGSI and web services are hoped to resolve
this - Dont need the whole thing
- Want to use GSI, need to install all of Globus
- COGs helped eliminate this burden
- Use only what you need
26Issues
- Globus originally seemed focused on compute grids
- Other groups were building data grids
- data grids represent a larger customer base
- GridFTP was over extended
- Was once proposed that it be used for more than
file transfers - Message passing
- VO Management
27A few of the Grid Projects with strong HEP
collaboration
Many national, regional Grid projects
-- GridPP(UK), INFN-grid(I), NorduGrid, Dutch
Grid,
European projects
US projects
Slide Taken From 8
28 The Large Hadron Collider Project 4 detectors
CMS
ATLAS
Storage Raw recording rate 0.1 1
GBytes/sec Accumulating at 5-8
PetaBytes/year 10 PetaBytes of
disk Processing 200,000 of todays fastest
PCs
LHCb
Slide Taken From 8
29Multi-Tiered View of LHC Computing
2.5-10Gbs
2.5-10Gbs
1-10Gbs
Slide Taken From 8
30Jlab Work
- Storage Resource Manager (SRM)
- Collaboration with LBL and FNAL
- Standard interface to a mass storage system
- Jasmine, Enstore, HPSS
- Uses
- GSI for authentication
- Web Services (before OGSA)
31Jlab Work
- Replica Catalog
- Web Services (before OGSA)
- Does not use GSI
- Mysql database used to store the map
- GFN -gt SURL1 SURLN
- Only one server/service
- May deploy mirrors of the database and proxies
- User Job Description Language (uJDL)
- Collaboration with BNL
- High level description of a physics job
32Jlab Work
- Job Scheduling Project
- Accepts a job description from a user
- Creates on or more jobs for execution
- If required, moves all necessary data to the site
where execution will take place - Moves output files to specified site, location
- Will use the uJDL but need to also define a
System Job Description Language - Lower level description of a compute job
33Job Scheduler Design
34Current Replica Catalog
Site A
Site B
3
2
1
4
5
6
8
7
Site C
Site D
35Proposed Replica Catalog Project
Site A
Site B
2
3
4
1
5
6
7
P2P Overlay Network
8
10
9
Site C
Site D
36Another Replica Catalog Problem
- Given
- Multiple VOs want to share data files
- Each VO represents a country or geographic area
- Each VO will maintain their own Replica Catalog
- Problem
- All VOs that want to participate need seamless
access to all Replica Catalogs - Individual VOs need to be able to come and go as
they please
37Another Replica Catalog Problem
USA
Canada
P2P Overlay Network
Europe
Japan
38References
- 1 Ian Foster. What is the Grid? A Three Point
Checklist. July 20, 2002. - 2 Ian Foster and Carl Kesselman, editiors. The
Grid Blueprint for a New Computing
Infrastructure. 1998. - 3 Ian Foster and Carl Kesselman. The Globus
Project A Status Report. Proceedings of
IPPS/SPDP98 Heterogeneous Computing Workshop.
1998. - 4 Ian Foster and Carl Kesselman. Globus
Ametacomputing Infrastructure Toolkit.
International Journal of Supercomputer
Applications. 1997 - 5 Globus Project. Globus Toolkit 2.2 MDS
Technology Brief. January 30, 2003. - 6 Globus Project. Community Authorization
Service (CAS) Overview. December 18, 2001. - 7 Ian Foster. Open Grid Services Architecture.
Plenary Talk at CHEP 2003. March 26, 2003. - 8 Ian Bird. Deploying the LHC Computing Grid.
Plenary Talk at CHEP 2003. March 26, 2003.