1

Observations on Architecture,Protocols, Services, APIs, SDKs, and the Role of the Grid Forum

Ian FosterWith: Carl Kesselman, Steven TueckeThanks also to: Bill Johnston, Marty

Humphrey, Rusty Lusk, Reagan Moore, and others

2

Overview

1. The Grid problem: controlled resource sharing in multi-institutional settings

2. Standards as a means of enabling sharing of code, resources, services

3. Aside: definition, role, and importance of protocols, services, SDKs, APIs, etc.

4. A “Grid Architecture”: a categorization of protocols, services, SDKs, and APIs

5. Questions for the Grid Forum

3

The Grid Problem

Grid R&D has its origins in high-end computing & metacomputing, but…

In practice, the “Grid problem” is about resource sharing & coordinated problem solving in dynamic, multi-institutional virtual organizations– Lack of central control, omniscience, trust

Primary challenge: to enable, maintain, and control the sharing of resources to achieve a common goal

4

Examples of Virtual Organizations

Members of a scientific collaboration– E.g., NSF PACIs, IPG, NEESgrid, GriPhyN– Sharing: computers, storage, software, …

Application server provider + customers– Sharing: ASP computers

Participants in peer-to-peer network– E.g., Gnutella, Napster, Entropia, …– Sharing: resources on individual PCs

Tremendous variety in scope, timescale, types of sharing, etc.

5

Universal Nature of the Grid Problem

“Sharing” fundamental in many settings– Application Service Providers, Storage Service

Providers, etc.; Peer-to-peer computing; Distributed computing; Business to business; …

Sharing issues not adequately addressed by existing technologies– Sharing at a deep level, across broad ranges

of resources and in a general way– E.g., user provides ASP with controlled access

to their data on an SSP: how?? Grid community has unique experience

6

Creating Usable Grids:What are the Challenges?

Approaches to problem solving– Data Grids, distributed computing, peer-to-

peer, collaboration grids, … Structuring and writing programs

– Abstractions, tools Enabling resource sharing across

distinct institutions– Resource discovery, access, reservation,

allocation; authentication, authorization, policy; communication; fault detection and notification; …

7

What is the Role of Grid Forum in Enabling Grid Computing?

1. Information exchange, of course Experiences, patterns, structures Useful even if every application & Grid is a

vertical “stovepipe”

2. Advocacy3. Enabler of shared effort

In code development: libraries, tools, … Via resource sharing: shared Grids In infrastructure

Opinion: Long term, only the third is sufficiently compelling to justify GF

8

Q: How do we Enable Shared Effort?A: “Standards” are Required

To enable portability/sharing of code– E.g., MPI lets me write portable // programs

To enable resource sharing– E.g., IP lets my computer speak to yours

To enable shared infrastructure– E.g., X.509 lets me share Certificate Authorities

But what sorts of “standards”?– Variously, APIs/SDKs, protocols, syntax, …– Observe that these are sometimes confused, so

let’s spend some time on definitions …

9

Some Important Definitions

Resource Network protocol Network enabled service Application Programmer Interface (API) Software Development Kit (SDK) Syntax

Not discussed, but important: policies

10

Resource

An entity that is to be shared– E.g., computers, storage, data, software

Does not have to be a physical entity– E.g., Condor pool, distributed file system, …

Defined in terms of interfaces, not devices– E.g. scheduler such as LSF and PBS define a

compute resource– Open/close/read/write define access to a

distributed file system, e.g. NFS, AFS, DFS

11

Network Protocol

A formal description of message formats and a set of rules for message exchange– Rules may define sequence of message

exchanges– Protocol may define state-change in

endpoint, e.g., file system state change Good protocols designed to do one thing

– Protocols can be layered Examples of protocols

– IP, TCP, TLS (was SSL), HTTP, Kerberos

12

Network Enabled Services

Implementation of a protocol that defines a set of capabilities– Protocol defines interaction with service– All services require protocols– Not all protocols are used to provide

services (e.g. IP, TLS) Examples: FTP and Web servers

Web Server

IP Protocol

TCP Protocol

TLS Protocol

HTTP Protocol

FTP Server

IP Protocol

TCP Protocol

FTP Protocol

Telnet Protocol

13

Application Programmer Interface

A specification for a set of routines to facilitate application development– Refers to definition, not implementation– E.g., there are many implementations of MPI

Spec often language-specific (or IDL)– Routine name, number, order and type of

arguments; mapping to language constructs– Behavior or function of routine

Examples– GSS API (security), MPI (message passing)

14

Software Development Kit

A particular instantiation of an API SDK consists of libraries and tools

– Provides implementation of API specification

Can have multiple SDKs for an API Examples of SDKs

– MPICH, Motif Widgets

15

Syntax

Rules for encoding information, e.g.– XML, Condor ClassAds, Globus RSL– X.509 certificate format (RFC 2459)– Cryptographic Message Syntax (RFC 2630)

Distinct from protocols– One syntax may be used by many protocols

(e.g., XML); & useful for other purposes Syntaxes may be layered

– E.g., Condor ClassAds -> XML -> ASCII– Important to understand layerings when

comparing or evaluating syntaxes

16

A Protocol can have Multiple APIsE.g., TCP/IP

TCP/IP APIs include BSD sockets, Winsock, System V streams, …

The protocol provides interoperability: programs using different APIs can exchange information

I don’t need to know remote user’s API

TCP/IP Protocol: Reliable byte streams

WinSock API Berkeley Sockets API

Application Application

17

An API can have Multiple ProtocolsE.g., Message Passing Interface

MPI provides portability: any correct program compiles & runs on a platform

Does not provide interoperability: all processes must link against same SDK– E.g., MPICH and LAM versions of MPI

ApplicationApplication

MPI API MPI API

LAM SDK

LAM protocol

MPICH-P4 SDK

MPICH-P4 protocol

TCP/IP TCP/IPDifferent message formats, exchange

sequences, etc.

18

Back to Grids:The Programming & Systems Problems

Approaches to problem solving– Data Grids, distributed computing, peer-to-

peer, collaboration grids, … Structuring and writing programs

– Abstractions, tools Enabling resource sharing across

distinct institutions– Resource discovery, access, reservation,

allocation; authentication, authorization, policy; communication; fault detection and notification; …

Programming Problem

Systems Problem

20

Aspects of the Programming Problem

Need for abstractions and models to add to speed/robustness/etc. of development– E.g., OO abstractions, MPI for messaging

Need for code/tool sharing to allow reuse of code components developed by others– E.g., MPI allows reuse of message passing– E.g., standard profilers, debuggers

Primary need is for standard programming environments: APIs and SDKs

21

Aspects of the Systems Problem

Need for interoperability when different groups want to share resources– Diverse components, policies, mechanisms– E.g., standard notions of identity, means of

communication, resource descriptions Need for shared infrastructure services to

avoid repeated development, installation– E.g., one port/service for remote access to

computing, not one per tool/application– E.g., Certificate Authorities: expensive to run

Need standard protocols, services, syntax

22

I.e., Standard APIs and Protocols are Both Important: For Different Reasons

Standard APIs/SDKs are important– They enable application portability– But w/o standard protocols, interoperability

is hard (every SDK speaks every protocol?) Standard protocols are important

– Enable cross-site interoperability– Enable shared infrastructure– But w/o standard APIs/SDKs, application

portability is hard (different platforms access protocols in different ways)

23

Grid “Architecture”

We now proceed to analyze Grid systems with respect to standards

Identify key areas where protocols, services, APIs, and SDKs can occur

Result is a layered protocol architecture

We assert this can be useful as a means of describing and structuring Grid Forum activities

24

Layered Grid Architecture(By Analogy to Internet Architecture)

Application

Fabric“Controlling things locally”: Access to, & control of, resources

Connectivity“Talking to things”: communication (Internet protocols) & security

Resource“Sharing single resources”: negotiating access, controlling use

Collective“Managing multiple resources”: ubiquitous infrastructure services

User“Specialized services”: user- or appln-specific distributed services

InternetTransport

Application

Link

Inte

rnet P

roto

col

Arch

itectu

re

25

Protocols, Services, and InterfacesOccur at Each Level

Languages/Frameworks

Fabric Layer

Applications

Local Access APIs and Protocols

Collective Service APIs and SDKs

Collective ServicesCollective Service Protocols

Resource APIs and SDKs

Resource ServicesResource Service Protocols

User Service ProtocolsUser Service APIs and SDKs

User Services

Connectivity APIs

Connectivity Protocols

26

An Aside on Terminology

Is this an “architecture” or just a “categorization” or “taxonomy”?– A matter of opinion (c.f. IAB: “Many

members of the Internet community would argue that there is no architecture”)

– Our opinion: it is somewhere in between, but is useful regardless

Becomes more architectural if/as we define “necessary” pieces at each level

Note that protocols says nothing about SDKs/APIs architecture (& vice versa)

27

Important Points

We build on Internet protocols– Communication, routing, name resolution, etc.

“Layering” here is conceptual, does not imply constraints on who can call what– Protocols/services/APIs/SDKs will, ideally, be

largely self-contained– But some things are fundamental: e.g.,

communication and security– But, advantageous for higher-level functions to

use common lower-level functions

28

ComputeResource

SDK

API

AccessProtocol

SourceCode Repository

SDK

API

LookupProtocol

Example: User Portal

Web Portal

Source code discovery, application configuration

Brokering, co-allocation, certificate authorities

Access to data, access to computers, access to network performance data

Communication, service discovery (DNS), authentication, authorization, delegation

Storage systems, schedulers

User

Appln

Collective

Resource

Connect

Fabric

29

ComputeResource

SDK

API

AccessProtocol

CheckpointRepository

SDK

API

C-pointProtocol

Example:High-Throughput Computing System

High Throughput Computing System

Dynamic checkpoint, job management, failover, staging

Brokering, certificate authorities

Access to data, access to computers, access to network performance data

Communication, service discovery (DNS), authentication, authorization, delegation

Storage systems, schedulers

User

Appln

Collective

Resource

Connect

Fabric

30

Standards, Again:Intergrid Protocols and Grid APIs

One or many protocols?– No one “right” protocol for any one function– But: interoperability requires that we define

and commit to core “Intergrid” protocols– Definition: “A resource is Grid-enabled if it

speaks Intergrid protocols” One or many APIs and SDKs?

– Many APIs, SDKs, programming models can target Intergrid protocols

– But: code sharing requires standards– So, e.g., “standard Grid collaboration APIs”

31

Questions for the Grid Forum

Is the “Grid architecture” described here a useful framework?– Could it be made more useful?– Are there things that it fails to capture or

misrepresents? Would it be a useful discipline for us to

try to place GF efforts in this context– E.g., be clear whether we are defining a

protocol, service, API, SDK, syntax (or something else: which is fine, too)

– E.g., explain (and argue about) where in the stack different pieces fit

32

Questions for the Grid Forum

Are some things easier, or more important, to standardize than others?– Protocols vs. APIs vs. syntax– Connectivity vs. resource vs. collective vs. user

layer protocols/services/APIs/SDKs I would suggest that

– Items lower in the stack tend to have broader impact, but standards useful at all levels

– Size of community effected (e.g., number of adopters) is the key figure of merit

– We should ask explicitly for such an analysis as part of a WG charter

33

Questions for the Grid Forum

Can we define core “intergrid protocols”?– I.e., instantiate (lower) layers in the diagram– We have avoided it until now (implies choice)– Until we do, interoperability is difficult

Possible approaches– Avoid seeking consensus, instead standardize

where it makes sense and where we can; rely on sense of “best practice” emerging

– Or, create an architecture WG, charged with defining requirements for “core protocols”??

– I think latter is better, unsure if it can work

34

Summary

Grids are about [large-scale] sharing– Hence require standard protocols to enable

interoperability and shared infrastructure– And, of course standard APIs and SDKs to

enable portability & code sharing– Both important; but very different

Well defined architecture can help understanding & progress– Provides a framework for figuring out where

the pieces fit– Facilitates asking questions such as “where

are standards particularly important?”

35

Questions?