National Computational Science University of Illinois at Urbana-Champaign Grid Forum 5 – October 2000 Overview of the Alliance Virtual Machine Room (VMR)

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

Overview of the Alliance Virtual Machine Room (VMR)

John TownsDivision Director, Scientific Computing

NCSA and the Alliancejtowns@ncsa.edu

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

Motivation

• Alliance mission is to prototype next-generation computational infrastructure for computational science and engineering community – Capability computing resources for very large-scale simulation

and data-processing applications– Capacity compute resources for loosely coupled computations

required massive compute cycles• Centralized compute facilities will not be able to satisfy all the

needs – Breakthrough simulations requiring more processing power

than available at a single site– Applications requiring multiple resources not available at single

site– Throughput requirements for servicing entire community

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

The Alliance VMR

• An evolving, persistent Alliance computational grid• Enable science by seamlessly integrating distributed

resources into a single computational environment

Give the User the PerceptionOf ONE Machine Room

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

What is the Alliance Virtual Machine Room?

• Infrastructure– Supercomputers,

networks, visualization resources, data archives and databases, instruments, etc.

• Middleware– Primarily Globus

components

• Grid services– Security infrastructure,

grid information sources, resource management, job submission an control, data management, etc.

• Portal interfaces and portal services– User Portal, Chemical

Sciences Portal, etc.

• Grid user support services– Consulting and helpdesk

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

Alliance Computational Resources at a Glance

• NCSA– SGI Origin2000

– 1536 processors

– HP X-Class (Exemplar)– 64 processors

– NT Supercluster– 256 processors

• Boston University– SGI Origin2000

– 192 processors

• Univ of Kentucky– HP N-Class cluster

– 96 processors

• Univ of New Mexico– RoadRunner cluster

– 128 processors

– Los Lobos cluster– 512 processors

• MHPCC– IBM SP

– 400+ processors

• Univ of Wisconsin– Condor flock

– 1200+ workstations

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

VMR Deployment Areas

• VMR Operations– Distributed operations support – Policies and Procedures

• Scheduling– Global scheduling of jobs

adhering to local constraints• Storage

– Integrate archive resources and add new capabilities

• Account Management– Account creation/management– Usage reporting for allocated

projects• Information Services

– Information services within Globus infrastructure

• Grid Security Infrastructure– PKI/GSI authentication

services– Interface to local policies and

mechanisms• Globus Installation and

Maintenance– Deploy Globus components

• User Services– Focus on supporting leading

edge applications/ research• User Portal

– User interface to VMR– Portal services to be leveraged

for specific research projects

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

VMR Operations

• VMR Resource Monitoring– Critical resource

information monitored by central VMR site management

– Tools and mechanisms to monitor system resources

• 24x7 Operations– Management policies and

procedures– Central VMR web site

• Common Helpdesk– Central VMR trouble ticket

system

• Base System Documentation and System Admin Support– Links to local system

documentation for each participating site

– System admin policies and procedures

• VMR Systems Software Repository– Current set of software

necessary for a site to participate in the VMR

– Developing VMR “tarball”

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

Scheduling

• Provide global scheduling of VMR resources for any users job– Must adhere to local policies and constraints– Allow for co-scheduling of resources at multiple

physical sites

• Provide global queuing system for user interface– Interfaces to local queuing systems

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

Account Management

• Infrastructure– Transaction-based info exchange

– Maintain local identities

• Account/Allocation management– Alliance Distinguished Name (DN) creation

– Account creation/removal centrally managed

• Usage reporting– Regular reporting from all sites of usage against

Alliance allocations

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

Grid Security Infrastructure

• Electronic credentials to serve as an Alliance Identity– Authentication to Alliance Resources– Support Single Sign-On – Used to establish confidentiality– Could be used to generate digital signatures

• Alliance PKI Certificate Authority– Certificate request, creation, expiration– Certificate management

• GSI deployment– Globus, GSI FTPd and sshd– Admin guide– Client deployment

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

Alliance User Portal aka MyGrid

• At SC99 prototype was shown

• Recent work has focused on component technologies– Applicable to other portal efforts!

• Also working on component applications

• Working with SDSC/NPACI and NASA/IPG to leverage efforts

• User Portal intended to be interface to VMR computing environment

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

Component Technologies

• MyProxy security– Credential delegation

and authentication for actions on behalf of the grid user

– Focus on authentication

• File transfer facilities – Primary concern is

moving (large) data files securely

– Java interface using GSIFTP

– Initial Java bean interface using GSIFTP

• Job submission – Mostly app specific

today– Initial general framework

prototype using plug-ins developed

• Search engine and documentation– Bought AltaVista license– Currently indexing all

HPC documentation at all partner sites

– Using this to re-vamp NCSA HPC documentation

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

Component Applications

• Systems/job status information – What is the state of every

machine in the VMR?– What is the state of every

job in the VMR?– Definition of XML DTD

formats– Currently implemented for

Origin2000• Mass store status

– Usage statistics– Network link status– Currently implemented for

NCSA archive

• Consulting access– On-line help with desktop

sharing + phone call– Trying out WebEx

• Allocations data access– Interface to check

allocation status from User Portal

– Direct access to centralized database backend

• Proposal process– Interface to manage

proposal process from User Portal

Grid Forum 5 – October 2000

National Computational ScienceUniversity of Illinois at Urbana-Champaign

Alliance User Portal Today