This is a static archive of the previous Open Grid Forum GridForge content management system saved from host forge.ogf.org file /sf/wiki/do/viewPage/projects.pgi-wg/wiki/RequirementsTable at Thu, 03 Nov 2022 00:04:34 GMT SourceForge : View Wiki Page: RequirementsTable

Project Home

Tracker

Documents

Tasks

Source Code

Discussions

File Releases

Wiki

Project Admin
Search Wiki Pages Project: pgi-wg     Wiki > RequirementsTable > View Wiki Page
wiki2239: RequirementsTable

PGI Requirements Table for agreement


Some of the sources of the collected requirements entered into the table :

  1. Existing OGF recommendations :  Mainly GLUEBESJSDLHPC Basic Profile
  2. PGI Security Model
  3. PGI : Execution Service Context - Main types of Grid Jobs
  4. Data Staging - Concepts and Scenarios
  5. Use case: GROMACS
  6. PGI Single Job State Model (available as ZARGO, XMI and PNG)
  7. PGI Job Vector Quick Submission Sequence Diagram (PNG)
  8. PGI Job Vector Submission With Full Validation Sequence Diagram (PNG)
  9. Strawman functionality
  10. Draft Specification Discussion Slides
  11. Open Issues
  12. PGI Execution Service Specification
  13. Background information from RFP of the US XD programme, NSF 08-571


This Table of Requirements has following columns :

  1    Nb    Requirement Number
  2    Description    Requirement Description
  3    Source    Source or Author of the Requirement
  4    Areas    Areas to which the requirement apply
  5    Dependencies    List of Requirement Numbers on which the current requirement depends
  6    Status    Requirement Status
  7    Date    Date of Status change (yyyy-mm-dd)


The Requirement Status can have one of following values :

  • Proposed   (Not yet validated as in scope and open for detailed discussion)
  • Rejected   (Not understandable, Out of scope, Contains hidden prerequisites, ...)
  • Dupl.   (Duplicate, therefore Closed)
  • Premature (Detailed specification, Specification of implementation, ...)
  • Open   (In scope, and under detailed discussion)
  • Agreed YES
  • Agreed NO
  • Met (The Wiki page can contain a link to the document and section)


Each requirement, if necessary, can point to a separate Wiki page providing :

  • The origin of the requirement,
  • The detailed description of the requirement,
  • Minutes of discussions
  • Minutes of agreement or rejection


Vocabulary and Terminology

The dedicated page PGI Vocabulary and Terminology details many terms. Below are presented only a few terms which are critical for the understanding of the requirements.

  • 'Activity' and 'Job' are fully synonyms.  'Job' is used by the current version of JSDL.  Since 'Activity' is used by the current versions of GLUE and BES, we will often use 'Activity' .
  • 'Client' :  Holder of credentials belonging to a member of a GLUE UserDomain.  For example, a Client MAY submit an Activity to the Execution Service, or query the Status of an Activity.
  • 'Payload' :  Anything (Application, Script, Pilot Job, ...) executed by the Computing Resource on request of the Activity.  The payload MAY completely ignore that is is executed inside a grid Activity.


NF)  Requirements concerning Non-functional areas (2010-03-26: 20 Req.)

Nb ID Description Source Areas Dependencies Status Date
NF1   Specifications MUST be designed so that the implementation and validation of Clients is easy Etienne Urbah Implementation issues, Cost   Dropped (Rejected, General Computer Science) 2010-04-27
NF1.1   Whenever multiple protocols are possible, Clients MAY implement only one Etienne Urbah Implementation issues, Cost NF1 Dropped (Rejected, Out of scope) 2010-04-27
NF1.2 161 Whenever multiple protocols are possible, Services SHOULD implement at least the 2 most used protocols Etienne Urbah Implementation issues, Cost NF1 Open 2010-04-28
NF2 145 Specifications MUST reuse as many existing standards as possible General Reuse Specs, Cost   Agreed YES 2010-04-28
NF3 146 Software components (Services and Clients) MUST ensure traceability of the original author of a request Etienne Urbah Event reports, Resilience, Accounting   Agreed YES 2010-05-10
NF4 147 Software components (Services and Clients) MUST generate adequate logs General Event reports, Resilience   Agreed YES 2010-04-28
NF5 148 Software components (Services and Clients) MUST generate and propagate meaningful error messages, including context description General Event reports, Resilience   Agreed YES 2010-04-28
NF6 162 Software components (Services and Clients) MUST implement the Security Policies defined by the JSPG (Joint Security Policy Group), in order to be resilient to attacks Etienne Urbah Resilience NF3, NF4 Open 2010-04-28
NF7 149 Specifications SHOULD prevent the occurrence of SPOFs and bottlenecks General Resilience, Availability, Performance   Agreed YES 2010-04-28
NF8 150 Clean specifications and interface definitions, don't keep obsolete and never used elements (by whom?) Balazs Konya Extensibility   Agreed YES 2010-04-28
NF9 151 The specification should be as much self-contained as possible, limit external dependencies Balazs Konya Implementation issues Conflicts with NF10, NF11, NF12, NF13 Open 2010-04-28
NF10 152 Evolutionary aspect of existing specifications (marketing efforts, bad community reputation if OGF standards are unstable).
Any new specification should be result of evolution of previous specification.
E.g. the scope and orientation of a spec should not change dramatically to the previous one, e.g. conceptual backward compatibility is a must.
NSF 08-571 Reuse Specs Conflicts with NF9 Agreed YES 2010-05-10
NF11 153 Backwards compatibility (on-the-wire) not a precise requirements, conceptual backward compatibility a must - new portType - new compilation anyway but general concept should be kept. NSF 08-571 Reuse Specs Conflicts with NF9 Open (unclear) 2010-04-29
NF12 154 Basic SW Engineering basic principles - implementation encapsulation, separation of policies, construction by composition, don't re-invent new mechanisms when some existing NSF 08-571 Reuse Specs Conflicts with NF9 Agreed YES 2010-04-28
NF13 160 Execution Service MUST NOT try to provide ALL grid functionalities, but MUST have a well defined scope, and MUST work together with other grid services :  Security,  Accounting,  Data (Storage, Movement, Catalog, ...),  perhaps Information,  Logging and Bookkeeping Etienne Urbah Reuse Specs, Cost Conflicts with NF9 Agreed YES 2010-05-20
NF14 155 Data is equally important as compute NSF 08-571     Open (unclear) 2010-04-29
NF15 156 Need to provide high availability and reliability NSF 08-571 Availability, Resilience Related to NF7 Agreed YES 2010-05-10
NF16 157 Single global namespace for resources of all kind (storage devices, compute device), for example: a path mechanism - 'address the same things everywhere'.  Definitely not XML namespaces - more identification of resources.  Closest analogy might be DNS. NSF 08-571 Practical Client Interface   Open (unclear) 2010-04-29
NF17 158 Easier usage to get more users involved. How does it effects PGI?!  A general rule like global namespaces (for example: path) NSF 08-571 Practical Client Interface   Open (unclear) 2010-04-29
NF18 159 Specifications of the Execution Service should be so well designed that its interfaces are likely to be re-used by other Services
(e.g. potential re-use of portTypes by other services - for example: information portType)
GIN Reuse Specs, Extensability   Agreed YES 2010-05-20


IS)  Requirements applying mainly to the Information functionality (2010-03-26: 18 Req.)

Nb ID Description Source Areas Dependencies Status Date
IS1 1 All grid entities (if possible) MUST be described using the GLUE model.  If not possible, extensions for the GLUE model are necessary GIN Information, Reuse Specs   Agreed YES 2010-06-22
IS1.1   All grid entities SHOULD be described in conformance with the latest published GLUE recommendation (currently GLUE 2.0) Etienne Urbah Information IS1 Re-proposed 2010-11-17
IS1.2 2 For each grid entity which can NOT be described in conformance with GLUE, PGI should propose extensions GIN Information IS1 Dropped (Dupl. IS1) 2010-04-29
IS2 163 Each Service MUST publish its properties (Name, Type, Endpoints, ...) and the properties of its Endpoints in conformance with GLUE recommendations Etienne Urbah Information   Re-proposed 2010-11-17
IS2.1   Each Service MUST publish the static and dynamic properties of each of its Endpoints to the Information functionality in conformance with GLUE recommendations Etienne Urbah Information IS2 Dropped (Dupl. IS2) 2010-04-21
IS2.2 3 A GLUE Computing Endpoint can describe only 1 Job submission interface (for example BES).
Job submission interfaces which have different properties (for example vector limits) MUST have different GLUE Computing Endpoints
Etienne Urbah Information IS2 Agreed YES 2010-06-22
IS3 4 Each service MUST publish information regarding its service properties GIN Information   Agreed YES 2010-04-28
IS4 5 The Execution Service MUST NOT expose detailed information about the GLUE entities which the Execution Service does not manage (all that are not expressed by the computing part of GLUE).  For example, Storage Element GLUE entity NOT exposed by Execution Service, NO details about Storage entity. Etienne Urbah Information, Activity Management NF13 Agreed YES 2010-04-28
IS5 6 Instance of Execution Service MUST provide access to attributes/meta data that describe this instance using GLUE2. (V-2 note: The appropriate sections of the XML rendering of GLUE2 need to be identified.). Different levels of verbosity should be offered. Strawman functionality Information   Agreed YES 2010-05-06
IS6 7 The Execution service instance MUST publish all data-staging capabilities (for example: supported protocols like GridFTP, ByteIO,...). Draft Spec Information, Activity Management, Data Management   Open 2010-05-05
IS7 8 If the Execution Service provides vector functionalities, then it MUST publish the vector limits as properties of GLUE Endpoints.
If these vector functionalities have different vector limits, then they MUST be grouped inside separate GLUE Endpoints entities having each a homogeneous vector limit.
Draft Spec Information, Activity Management IS2.2 Agreed YES 2010-04-28
IS8 164 For not yet finished Activities, the Information functionality MUST accept requests querying the Activity Status, and MUST return this Status with different levels of verbosity (basic, detailed, more detailed) Etienne Urbah Information, Activity Management   Open (unclear) 2010-04-28
IS9 165 For already finished Activities, the Information functionality SHOULD accept requests querying the Activity Status, but MAY return an error Etienne Urbah Information, Activity Management   Agreed YES 2010-05-20
IS10   For requests querying Activity Status, the Information functionality MUST accept, as input parameter, 1 Activity ID Etienne Urbah Information, Activity Management IS8 Re-proposed 2010-05-05
IS11   For requests querying Activity Status, the Information functionality MUST accept, as input parameter, a vector of Activity IDs Etienne Urbah Information, Activity Management IS8 Dropped (Rejected, Bad prerequisite on Information Service) 2010-04-21
IS12   For requests querying Activity Status, the Information functionality MUST accept, as input parameter, complex queries on Activities expressed according to the GLUE2 model without explicit specification of the Activity IDs.  The Information functionality MAY support multiple query languages in addition to the mandatory one. Etienne Urbah Information, Activity Management IS8 Re-proposed 2010-05-05
IS13 166 For requests querying Activity Status, the Information functionality SHOULD support query languages Etienne Urbah Information, Activity Management IS8 Agreed YES 2010-05-20
IS14 172 In order to prevent overloading the Execution Service (which has performance requirements for Activity Management), the Information functionality SHOULD be separated from the Execution Service Etienne Urbah Information, Execution Service NF13, IS5, IS12, IS13 Agreed YES 2010-05-20


AA)  Requirements applying mainly to Security (Authentication, Authorization, Delegation) (2010-03-26: 31 Req.)

Nb ID Description Source Areas Dependencies Status Date
AA1 9 If a server authenticates itself to clients, then it MUST do so with X509 certificates on TLS.  It MAY optionally use other mechanisms. General Security   Agreed YES 2010-06-22
AA1.1 10 SSL certificates of servers MUST be signed by a CA belonging to IGTF General Security AA1 Open 2010-04-28
AA2 11 Each Service MUST publish the Authentication and Authorization methods accepted by its Endpoints in conformance with GLUE recommendations General Security, Information IS2.1 Agreed YES 2010-06-22
AA2.1   The Authentication and Authorization methods accepted by each Endpoint must be published inside the 'Capability' Attribute of the 'Endpoint' entity of the Information Service, as extensions of the 'security.authentication' and 'security.authorization' values of the 'Capability_t' data type Etienne Urbah Security, Information AA2, IS2.2 Dropped (Premature, Implementation) 2010-04-21
AA3 12 For Client authentication, Services MUST require security credentials General Security   Open 2010-04-28
AA3.1 13 For Client authentication, Services MAY accept full X509 certificates General Security AA3 Open (unclear) 2010-04-28
AA3.2 14 For Client authentication, Services MAY accept RFC-3820-compliant X509 proxies General Security AA3 Open (unclear) 2010-04-28
AA3.3 15 For Client authentication, Services MAY accept SAML assertions General Security AA3 Open (unclear) 2010-04-28
AA3.4 16 For Client authentication, Services MAY accept Shibboleth credentials General Security AA3 Open (unclear) 2010-04-28
AA3.5 17 For Client authentication, Services must accept all following authentication methods: Full X509, X509 Proxy General Security AA3 Agreed YES 2010-06-22
AA4 18 For Client authorization, Services MUST require security credentials General Security   Open 2010-04-28
AA4.1 19 For Client authorization, the packaging of security credentials MUST be designed to prevent that any software component gains undue privilege by extracting only some credentials from the package Etienne Urbah Security AA4 Agreed YES 2010-04-28
AA4.2 20 For Client authorization, Services MAY accept the DN of the X509 certificate or RFC-3820-compliant X509 proxy GIN Security AA4 Agreed YES 2010-04-28
AA4.3 21 For Client authorization, Services MAY accept X509 VOMS-style Attribute Certificates (VOMS extensions) GIN Security AA4 Agreed YES 2010-04-28
AA4.4 22 For Client authorization, Services MAY accept SAML assertions GIN Security AA4 Agreed YES 2010-04-28
AA4.5 23 For Client authorization, Services MAY accept Shibboleth credentials GIN Security AA4 Agreed YES 2010-04-28
AA4.6 24 For Client authorization, Services MUST accept at least one of DN, X509 VOMS extensions, SAML assertions GIN Security AA4.2, AA4.3, AA4.4 Open 2010-04-28
AA4.7 25 For Client authorization, Services SHOULD accept SAML assertions GIN Security AA4.4 Open 2010-04-28
AA4.8 26 Cross-organization delegation, particularly to (university) campuses and other partner institutions, use native authentication mechanisms to national resources.
I can use my credentials at University X to retrieve another credential for University Y that is in the same trust federation.
For example: in US a lot of efforts towards SAML and 'incommon' driven by Internet2 (i.e. Shibboleth).
Different attributes: Ways to use WS-Trust to work on this.
NSF 08-571 Security AA4 Open 2010-04-28
AA5   For Client authentication and authorization, Services MUST follow the OGSA Basic Security Profile 2.0, which references Secure Addressing Profile 1.0 and Secure Communications Profile 1.0 Etienne Urbah Security Perhaps conflicts with AA6 Dropped (Rejected, NOT precise enough) 2010-04-21
AA6   For Client authorization, Services MUST implement the Common Open Policy Service (COPS) defined by RFC 4261, with PAP, PDP, PEP, ... Etienne Urbah Security Perhaps conflicts with AA5 Dropped (Rejected, NOT precise enough) 2010-04-21
AA7 27 In order to keep middleware complexity and bandwidth usage as low as possible, grid Services should NOT send the full description of their security interface inside each message, but only when specifically requested GIN Security, Performance   Open 2010-04-28
AA8 28 Service has to act on behalf of another identity.  Services often need to delegate those Credentials to other Service.  Therefore, Services MUST offer interoperable methods for Credential delegation. GIN Security   Agreed YES 2010-04-28
AA8.1   For credential delegation, Services MUST follow the XACML profile and implementation for Authorization Interoperability between OSG and EGEE (David Groep can explain) Etienne Urbah Security AA7; Perhaps conflicts with AA8.2 Dropped (Premature, Detailed specification) 2010-04-21
AA8.2   For credential delegation, Services MUST implement Use of WS-TRUST and SAML to access a Credential Validation Service, which is based on WS-Trust 1.3 Etienne Urbah Security AA7; Perhaps conflicts with AA8.1 Dropped (Premature, Detailed specification) 2010-04-21
AA8.3 29 We need delegation and have to define exactly what delegation is and its methods.
E.g. the delegation procedure might require multiple steps (one approach might be the delegation portType),
e.g. gridsite delegation method - bringing delegation to the portType level using operations.
GIN Security AA8 Agreed YES 2010-04-28
AA8.4 30 Delegation credential renewal should be supported GIN Security AA8 Agreed YES 2010-04-28
AA8.5 31 Associate Activity to a specific delegation credential, e.g. certificate or SAML assertion, or...
Each job belongs to a credential (associate them), when ever one credential expires then the Execution Service may terminate the corresponding Activities
Luigi Zangrando Security AA8 Agreed YES 2010-04-28
AA9 32 There must be a mechanism which allows users to manage Activities submitted by other users (access control lists/methods/policies).
In order to authorize (or not) an request on an Activity, each instance of the Execution Service MUST enforce a consistent authorization method.
Luigi Zangrando Security, Activity Management   Agreed YES 2010-04-28
AA10   Any action/invocation/functionality of the Execution Services requires authorization of end-users/clients   Security   Dropped (Dupl. AA4) 2010-04-21
AA11 33 It is assumed that the clients obtain security setup info through out-of-band mechanisms which are not covered by this specification. Draft Spec Security   Open 2010-04-28
AA12   Clients MUST authenticate Services with SSL certificates Luigi Zangrando Security   Dropped (Dupl. AA1) 2010-04-21


AR)  Requirements applying mainly to Application Repository (2010-03-26: 3 Req.)

Nb ID Description Source Areas Dependencies Status Date
AR1 34 An easy way of launching applications or pre-configured /pre-installed software w/o specifying location details.
Installed/pre-configured ones should be exposed as well as part of the resource description.
The available software should be exposed through the service and in turn be requestable for resource/applications statements in JSDL.
E.g. extending the JSDL with "pre-configured sw pieces" and/or "software libraries", e.g. global namespace of this applications might be beneficial as well. , etc.
GIN, NSF 08-571 Application Repository, Information, Job Description, Activity Management   Agreed YES 2010-04-29
AR2 35 Application management, for example: pre-installed applications, abstract notion of the application (i.e. specify w/o executable locations), global namespace of this applications might be beneficial as well. NSF 08-571 Application Repository, Job Description   Dropped (Duplicate AR1) 2010-04-29
AR3   Application management should use the Application Repository specified in the deliverable DSA1.1 of the EDGeS project Etienne Urbah Application Repository AR1 Dropped (Premature, Specification of implementation) 2010-04-21


Ac)  Requirements applying mainly to Accounting (2010-03-26: 2 Req.)

Nb ID Description Source Areas Dependencies Status Date
Ac0 167 In order to permit efficient Accounting, a grid infrastructure MUST provide an Accounting functionality (without real time requirements) which consolidates Accounting records at the grid level from various systems, securely keeps Accounting records even after Activities have finished, implements complex accounting queries, and supports the corresponding CPU load Etienne Urbah Accounting   Open 2010-04-29
Ac1   In order to prevent overloading the Execution Service (which has performance requirements for Activity Management), the Accounting functionality MUST be separated from the Execution Service Etienne Urbah Accounting, Activity Management Ac0, NF13 Re-proposed 2010-04-23
Ac2 36 The Execution Service SHOULD provide Accounting records for each managed Activity. e.g. OGF Usage Records, for tracking resource usage.
Most likely improving UR in terms of network and storage resource tracking and improvements of compute parts of multi-core-business.
Grid-level attributes (for example: start-time on Grid vs. in batch-system).
NSF 08-571 Accounting   Agreed YES 2010-04-29


LB)  Requirements applying mainly to Logging and Bookkeeping (2010-03-26: 8 Req.)

Nb ID Description Source Areas Dependencies Status Date
LB1 37 In order to permit efficient Log analysis and Security audit, some Logging and Bookkeeping functionality MUST secure persistence at the grid level of Activity logs from various logging systems and MUST permit Client access to these Activity logs, even after Activities have finished PGI Logging and Bookkeeping, Execution Service NF2, NF3, NF6 Re-proposed 2010-05-07
LB2 38 Some Logging and Bookkeeping functionality MUST accept requests querying the Status of an Activity (primarily for completed and finished Activities), and MUST return the Activity Status with different levels of verbosity (basic, detailed, more detailed) PGI Logging and Bookkeeping, Execution Service LB1 Re-proposed 2010-05-07
LB3 39 Some Logging and Bookkeeping functionality MUST accept requests in order to obtain the Activity log, and MUST return the Activity History with different levels of verbosity (basic, detailed, more detailed) PGI Logging and Bookkeeping, Execution Service LB1 Agreed YES 2010-04-29
LB4   For all requests, the Logging and Bookkeeping functionality MUST accept, as input parameter, 1 Activity ID Etienne Urbah Logging and Bookkeeping LB1 Re-proposed 2010-05-07
LB5   For all requests, the Logging and Bookkeeping functionality MUST accept, as input parameter, a vector of Activity IDs Etienne Urbah Logging and Bookkeeping LB1 Dropped (Premature, Detailed specification) 2010-04-21
LB6   For all requests, the Logging and Bookkeeping functionality MUST accept, as input parameter, complex queries on Activities expressed according to the GLUE model without explicit specification of the Activity IDs.  The Logging and Bookkeeping functionality MAY support multiple query languages in addition to the mandatory one. Etienne Urbah Logging and Bookkeeping LB1 Re-proposed 2010-05-07
LB7 40 In terms of queries, the Logging and Bookkeeping functionality SHOULD support query languages, e.g. at least one complex query language like XPath, XQuery, in order to obtain the activity logs, etc PGI Logging and Bookkeeping LB1 Open 2010-04-29
LB8 41 In order to prevent overloading the Execution Service (which has performance requirements for Activity Management), the Logging and Bookkeeping functionality SHOULD be separated from the Execution Service Etienne Urbah Logging and Bookkeeping, Execution Service NF13, LB1...LB7 Open 2010-04-29


JM)  Requirements applying mainly to Activity Management (2010-03-26: 67 Req.)

Nb ID Description Source Areas Dependencies Status Date
JM1 42 For Activity submission, Clients MUST query the Information Service in order to get an adequate Endpoint of an Execution Service Etienne Urbah Activity Management, Information Service   Proposed 2010-03-17
JM2 43 The Execution Service MUST provide a way to create Activities. General Activity Management   Proposed 2010-03-19
JM3 44 Activities submitted by Clients to an Execution Service MUST be specified using a well-defined Job Description document General Activity Management, Job Description JM2 Proposed 2010-03-17
JM4 45 On creation of an Activity, the Execution Service MUST return to the Client an Activity ID permitting the Client to perform subsequent actions (Query, Cancel, ...) on this precise Activity General Activity Management, Information Service JM2 Proposed 2010-03-17
JM4.1 46 The service assigns a global unique Activity ID - not the client General Activity Management, Information Service JM4 Proposed 2010-03-20
JM4.2   Potentially the WS-RF Factory pattern (BESFactory creates Activity Service instances, etc.) might be a good option and a good design in this context is the requirement.  Mabye running into performance issues with multiple/bulk vector operations.  It potentially makes it easier to use WS-Notification approach. NSF 08-571 Activity Management, Information Service JM4 Dropped (Premature, Detailed specification) 2010-04-23
JM4.3 47 The Activity ID MUST be suitable for manual transfer by a human (In case of problems with the Activity) Etienne Urbah Activity Management, Information Service JM4 Proposed 2010-03-17
JM4.4 48 In case of an Activity being moved to another Execution Service, the Activity MUST still be accessible via the initial Execution Service which created the Activity Etienne Urbah Activity Management JM4; Conflicts with JM4.5 and JM4.6 Proposed 2010-03-17
JM4.5 49 An Execution Service MAY offer several Endpoints to create and later manage the Activities while the later Endpoint must stay fixed Etienne Urbah Activity Management JM4; Conflicts with JM4.4 and JM4.6 Proposed 2010-03-17
JM4.6 168 After creation of the Activity, the Endpoint used by the Client to manage the Activity MAY vary Etienne Urbah Activity Management JM4; Conflicts with JM4.4 and JM4.5 Re-proposed 2010-04-27
JM4.7 50 The Execution Service MUST NOT assume that the Client is able extract any information from the Activity ID.  Therefore, the Activity ID can be completely opaque. General Activity Management JM4.4; Conflicts with JM4.8 and JM4.9 Proposed 2010-03-17
JM4.8 51 From the Activity ID, the Client MUST be able to extract the Endpoint suitable for Activity Management, but the Execution Service MUST NOT assume that the Client is able to extract any other information from the Activity ID General Activity Management JM4.5; Conflicts with JM4.7 and JM4.9 Proposed 2010-03-17
JM4.9 52 From the Activity ID, the Client MUST be able (for example by querying a Name Server) to dynamically retrieve the Endpoint suitable for Activity Management, but the Execution Service MUST NOT assume that the Client is able to extract or retrieve any other information from the Activity ID General Activity Management JM4.6; Conflicts with JM4.7 and JM4.8 Proposed 2010-03-17
JM4.10 53 Define exactly what an Activity ID is. GIN Activity Management, Information Service JM4.1, JM4.2, JM4.3, JM4.4, JM4.5, JM4.6, JM4.7, JM4.8, JM4.9 Proposed 2010-03-19
JM4.11 54 Operations dealing with Activity management of the service SHOULD operate on Activity IDs in order to ensure a good performance GIN Activity Management JM4.1, JM4.10 Proposed 2010-03-20
JM5 55 A simple Activity is defined by a simple Job Description containing only ONE local job executed by only ONE batch system Etienne Urbah Job Description, Activity Management JM2 Proposed 2010-03-17
JM5.1 56 The Execution Service SHOULD accept vectors of simple Activities as input parameter for several operations such as Create(), ChangeState(), ActivityStatus(), Cancel(), Purge() (implementation aspect longly discussed in Munich - worth to keep) Draft Spec Activity Management JM5 Proposed 2010-03-17
JM5.2 57 The execution service may have limits with respect to the size of the vector due to implementation aspects and if requests exceed the implementation limits the service communicates this via a fault. Draft Spec Activity Management JM5.1 Proposed 2010-03-20
JM5.3 58 The Execution Service will NOT take into account Collection Activities (these are NOT vectors), which are containers for a limited number of explicitly described independent Single Activities Draft Spec Activity Management JM5 Proposed 2010-03-17
JM5.4 59 The Execution Service will NOT take into account Parameter Sweep Jobs, which describe independent Single Activities to be created dynamically Draft Spec Activity Management JM5 Proposed 2010-03-17
JM5.5 60 The Execution Service will NOT take into account DAG Jobs, which describe a DAG workflow of a limited number of explicitly described Single Jobs Draft Spec Activity Management JM5 Proposed 2010-03-17
JM5.6 61 The Execution Service SHOULD support any Activity running massively-parallel processes using MPI and large-scale HPC Systems Strawman functionality Activity Management JM5 Proposed 2010-03-18
JM6 62 The Execution Service MUST log enough grid information inside logging systems (which MAY vary during the lifetime of the Activity), in order to permit efficient Log analysis and Security audit Etienne Urbah Activity Management, Event reports JM2 Proposed 2010-03-17
JM6.1 63 The Execution Service MUST accept requests querying the Status of an Activity, and MUST return the Activity Status Draft Spec Activity Management, Event reports JM6 Proposed 2010-03-17
JM6.2 64 The Execution Service MUST provide the Activity Status information with different levels of verbosity (basic, detailed, more detailed) Draft Spec Activity Management, Event reports JM6 Proposed 2010-03-18
JM6.3   The Logging & Bookkeeping Service, which consolidates Activity logs from various logging systems, MAY be completely separate from the Execution Service.  But the Logging & Bookkeeping Service MAY also be provided by the Execution Service, in which case the Execution Service MUST accept requests for the log of an Activity, and MUST return the Activity log. Etienne Urbah Activity Management, Event reports JM6 Dropped (Rejected, Bad prerequisite on Logging & Bookkeeping Service) 2010-04-23
JM6.4   In order to minimize polling of the Activity Status, the Execution Service MAY implement a mechanism for notification. Etienne Urbah Activity Management, Event reports, Notification JM6 Dropped (Dupl. JM6.5) 2010-04-23
JM6.5 65 In order to minimize polling of the Activity Status, the Execution Service MAY implement a mechanism for notification.  Asynchronous notification of Activity status (for example: e-mail, SMS, etc.) - human level notification NSF 08-571 Activity Management, Notification JM6 Proposed 2010-03-20
JM6.6   If notification in PGI, then OASIS WS-Notification (automatic notification) might be a good option.  BES couldn't decide on a notification mechanism, because too many vendors in the boat.  Experience in implementation in Grids present (for example: UNICORE) NSF 08-571 Activity Management, Notification JM6 Dropped (Premature, Specification of implementation) 2010-04-23
JM6.7 66 The Instance of Execution Service MUST permit any Client to request the list of all Activities managed by this instance on which the Client has authorization Strawman functionality Activity Management JM6 Proposed 2010-03-18
JM6.8 67 If a Client queries for an already finished Activity, the Execution Service MAY response with an error. Etienne Urbah Activity Management, Information Service JM6 Proposed 2010-03-26
JM6.9 68 The Execution Service MUST accept requests querying the detailed Information of an Activity, and MUST return the detailed Activity Information (more than just the status, for example: GLUE2 Activity).  Provide Activity information with different levels of verbosity (basic, detailed, more detailed). Draft Spec Activity Management, Information Service JM6 Proposed 2010-03-20
JM6.10 69 Obtain Activity session directory/staging-directory (maybe as part of more general) information via an Activity ID. Draft Spec Activity Management, Data Management JM4, JM6 Proposed 2010-03-19
JM6.11   For all requests concerning Activities, the Execution Service functionality MUST accept, as input parameter, 1 Activity ID   Activity Management, Information Service JM4, JM6 Re-proposed 2010-04-26
JM6.12 70 The Execution Service MAY provide the functionality to monitor (poll) the state of activities via filters (defined by queries, i.e. not requiring explicit Activity ID). Luigi Zangrando Activity Management, Information Service JM6 Proposed 2010-03-19
JM6.13 71 The Execution Service MUST provide information about the Activity history (i.e. Activity status) and invoked operations on this Activity (log on ws level?!). Adopt potentially the JSDL instance specification. Etienne Urbah Activity Management, Information Service JM6 Proposed 2010-03-20
JM6.14 72 The execution service MUST publish state information according to the common PGI state model, but status information might be optionally published according to multiple state models. Draft Spec Activity Management, Information Service JM6 Proposed 2010-03-20
JM6.15 73 The execution service should offer more flexibility enabling complex queries on resource and Activity information expressed according to the GLUE2 model without explicit specification of the Activity IDs.  The Execution Service MAY support multiple query languages in addition to the mandatory one. Draft Spec Activity Management, Information Service JM6 Proposed 2010-03-20
JM6.16 74 The execution service SHOULD offer separate(!) operations to query Resource and Activity information. Draft Spec Activity Management, Information Service JM6 Proposed 2010-03-20
JM6.17   An Execution Service mechanism in terms of queries should support Xpath 1.0 and MAY support additional query languages such as XQUERY, SQL, sparql and custom defined ones (for example: python-based).   Activity Management, Information Service JM6 Dropped (Premature, Specification of implementation) 2010-04-23
JM7   The Execution Service MUST provide enough Usage Records as specified by UR to the Accounting Service Etienne Urbah Activity Management, Accounting JM2 Dropped (Dupl. Ac2) 2010-04-23
JM7.1   The Accounting Service MUST be completely separate from the Execution Service.  Therefore specifications will NOT take into account Client requests concerning Accounting. Etienne Urbah Activity Management, Accounting JM7 Dropped (Dupl. AC1) 2010-03-17
JM8 75 In order to apply the principle of 'Full Automation of Processing', the Activity State Model SHOULD NOT have 'Hold' States Etienne Urbah Activity Management / State Model Conflicts with JM9 Proposed 2010-03-18
JM9 76 In order to permit the Client to perform client-directed processing of submitted Activities (for example client-directed data staging), the Execution Service SHOULD manage 'Hold' points for Activities.  Fault if not supported. Strawman functionality Activity Management / State Model Conflicts with JM8 Proposed 2010-03-18
JM9.1 77 The Execution Service MUST be able to create Activity in 'Hold' state Draft Spec Activity Management / State Model JM9 Proposed 2010-03-19
JM9.2 78 Requirement to have a pending state (i.e. waiting to be scheduled) Morris Riedel Activity Management / State Model JM9 Proposed 2010-03-19
JM9.3 79 When the client wants to perform client initiated data-staging-in, the client MUST also specify in the Job Description via 'Hold' points that the Activity holds. Draft Spec Activity Management / State Model, Job Description JM3, JM9 Proposed 2010-03-20
JM9.4 80 The Execution Service MAY offer the 'Hold' and corresponding 'Resume' functionality of Activity processing.  Change state operation to resume a hold execution and to put the Activity into 'Hold'. Draft Spec Activity Management / State Model JM9 Proposed 2010-03-20
JM10 81 On Activity submission, if the Execution Service fails to create the Activity for any reason, the Execution Service MUST return to the Client a descriptive error General Activity Management JM2 Proposed 2010-03-18
JM10.1 82 It is important that optional Job Description elements MUST NOT be ignored by the service and rather throw a corresponding fault in the case that the service does not support this optional client request. GIN Activity Management JM3, JM10 Proposed 2010-03-20
JM11 169 The Execution Service MAY provide its functionalities to Clients through various protocols (RPC, Corba, ActiveMQ, ...), but MUST provide them to Clients through a Web Service General Activity Management   Re-proposed 2010-04-26
JM11.1 83 The Execution Service MUST provide to the Client a Web Service using SOAP Etienne Urbah Activity Management JM11 Proposed 2010-03-18
JM12 84 The Execution Service is NOT REQUIRED to provide any service management capability (for example start/stop accepting new Activities). GIN Activity Management   Proposed 2010-03-19
JM13 85 Activity recovery via 'changestatus' mechanism after Activity was failed. Balazs Konya Activity Management   Proposed 2010-03-19
JM14 86 The Execution Service offers the possibility to 'cancel/terminate/kill/destroy' an Activity. The cancel means that active data-staging processes SHOULD be cancelled and active executions in RMS MUST be cancelled. Strawman functionality Activity Management, Data Management   Proposed 2010-03-19
JM15 87 Lease of Activities (several Activities may get purged by the Execution Service if they loose contact with the client or after a timeout), 'automatic cleanup' in case the lease was not renewed Luigi Zangrando Activity Management   Proposed 2010-03-19
JM16 88 Request routing capabilities (for example: preferred shares, etc.) - Prepare for TIME CONSUMING DISCUSSIONS GIN Activity Management   Proposed 2010-03-19
JM17 89 Request Forwarding capabilities - Prepare for TIME CONSUMING DISCUSSIONS GIN Activity Management   Proposed 2010-03-19
JM18 90 Execution Service SHOULD perform validation (potentially several steps).  An incomplete validation can be done immediately while a full validation will be only performed asynchronously, as requested by Client Draft Spec Activity Management JM4.1, JM4.10 Proposed 2010-03-19
JM18.1 91 The validation MUST happen as part of the Activity creation and not later.  If validation fails, an error has to be provided. Draft Spec Activity Management JM4.1, JM4.10 Proposed 2010-03-20
JM18.2 92 The Execution Service can decide which level of activity validation is appropriate GIN Activity Management   Proposed 2010-03-20
JM19 93 The execution service should offer Time estimations to completion of asynchronous operations, such as State change, Cancellation, Purge Draft Spec Activity Management   Proposed 2010-03-20
JM20 94 Requirement to have a purge (maybe called wipe) operation.  This operation is only allowed on any final state according to the state model.  The operation is not supposed to kill Activities, that is why its only allowed on final states.   Activity Management / State Model   Proposed 2010-03-20
JM21 95 Cleanup of BES - a number of compromises were made to satisfy parties, drop, improve, etc. - but not breaking the general BES concept (more on GridForge BES session) NSF 08-571 Non-functional   Proposed 2010-03-20
JM22 96 If underlying naming and binding mechanism be resilient to failure to migration and replication, for example: create Activity and its decided its to hand it of to another BES and (a) makes this new BES responsible or (b) migrate completely to the new BES (i.e. automatic forwarding).  Migration is an important activity not only in compute but also in data. NSF 08-571 Activity Management, Availability, Resilience   Proposed 2010-03-20
JM23 97 A user should be able to know the full routing of the forwarding actions of Activities by services.  JSDL instance is partly tackling this problem. NSF 08-571 Activity Management, Job Description   Proposed 2010-03-20
JM24 98 The user should be able to contact the end of the path (final) service. NSF 08-571 Activity Management   Proposed 2010-03-20
JM25 99 StartAccepting and StopAccepting activities might be still required (for example: for Desktop Grids and HTC).  Preserve backwards compatibility from PGI perspective. NSF 08-571 Service Management   Proposed 2010-03-20


JD)  Requirements applying mainly to Job Description (2010-03-26: 25 Req.)

Nb ID Description Source Areas Dependencies Status Date
JD1   Activities submitted by Clients to an Execution Service MUST be specified using a well-defined Job Description document Etienne Urbah Job Description, Activity Management   Dropped (Dupl. JM3) 2010-04-23
JD2   Specifications will NOT take into account JSDL Parameter Sweep Activity Extension Etienne Urbah Job Description, Activity Management   Dropped (Dupl. JM5.3) 2010-04-23
JD3 100 The Job Description document MUST reference all grid entities in conformance to the GLUE specification GIN Job Description, Information Service   Agreed YES 2010-04-28
JD4 101 The Job Description document specification MUST permit the Client to request automatic data stage-in and stage-out GIN Job Description, Activity Management   Agreed YES 2010-04-28
JD5 102 The Job Description document specification MUST permit the Client to request the number of times the Execution Service should perform local automatic resubmission in case of Activity failure Etienne Urbah Job Description, Activity Management   Agreed YES 2010-04-29
JD6 103 The Job Description document specification MUST permit the Client to request that at specified Activity states, the Execution service sets the Activity on 'Hold' Draft Spec Job Description, Activity Management JM9 Agreed YES 2010-04-28
JD7 104 The Job Description SHOULD allow to specify n different alternative(!) data sources of any single input data. (Semantics: data may be fetched from some of the alternative sources) Draft Spec Job Description, Data Management   Agreed YES 2010-04-28
JD8 105 Job Description supports :
1) Service-directed data-staging (i.e. specify data-stagings),
2) Client-initiated data-staging (i.e. specify data-staging but service will do nothing, only control whether files existing),
3) Manual data-staging (i.e. no specification of data-staging elements)
Draft Spec Job Description, Data Management, Activity Management JM9 Agreed YES 2010-04-28
JD9 106 The Job Description SHOULD allow to specify n different data targets of any single output data.  Later to be decided how the targets will be handled. Draft Spec Job Description   Open (unclear after 10 mn) 2010-04-28
JD10   Specify exactly what happens when one data-staging (sources and target) element can not be satisfied: complete failure? Or continue?   Job Description JD8, JD9 Dropped (Premature, Specification of Implementation) 2010-04-23
JD11 107 Job Description should be only used for Job Description not for any kind of feedback.  Activity status information should not be communicated with a 'changed JSDL'. Draft Spec Job Description   Agreed YES 2010-04-28
JD12   Job Description should be only used for Job Description, NOT for any kind of feedback.  The main status information should NOT be communicated with a 'changed JSDL'.  Job Description MUST be strictly used only for expressing Activity requirements Draft Spec Job Description   Dropped (Dupl. JD11) 2010-04-23
JD13 108 Job Description comes with support for requesting multi-processor Activities (for example: threads/node, network topology, task/core mappings, multi-threading and such like) Strawman functionality Job Description   Agreed YES 2010-04-28
JD14 109 Job Description document should be with simple structure, which avoids unnecessary overlaps (e.g. memory in different places defined) GIN Job Description   Open (NO agreement after 10 mn) 2010-04-28
JD15 110 The PGI Job Description document MUST provide a simple structure for ranges or a proper description of ranges, e.g. simpler than currently defined in JSDL Draft Spec Job Description   Open (unclear after 10 mn) 2010-04-28
JD16 111 Job Description document should offer grouped data-stagings for a better clean structure Draft Spec Job Description   Open (NO agreement after 10 mn) 2010-04-28
JD17 112 Job Description document should offer a clear separation between resource requirements and application requirements
(for example: suggest to put memory information into resource), e.g. improving the JSDL specification in this context (if necessary)
Draft Spec Job Description   Agreed YES 2010-04-28
JD18 113 Job Description document should offer a new re-usable structure to describe an executable composed as path and argument Draft Spec Job Description, Application Repository   Open (unclear after 10 mn) 2010-04-28
JD19 114 Job Description document should offer a new re-usable structure to describe a software requirement, to express relations (OR, AND, GREATER, SMALLER, ...) of software requests and other requests   Job Description JD24 Agreed YES 2010-04-29
JD20 115 Job Description document should offer a new re-usable structure to define 'scalabletime' requests optionally depending on benchmark results. Draft Spec Job Description   Open (unclear after 10 mn) 2010-04-29
JD21   Refinements and Tunings of numerous JSDL elements according to production experience.   Job Description   Dropped (Rejected, NOT precise enough) 2010-04-23
JD22 116 Clearly separate file and directory in the data-stagings., e.g. within the job description there the element are clearly labelled as file or directory , e.g. different aspect of handling directory or a file, you can not differentiate Draft Spec Job Description   Agreed YES 2010-04-29
JD23 117 New structure to describe the context in which the Job Description was generated (i.e. meta) Draft Spec Job Description   Agreed NO 2010-04-28
JD24 118 Improvement of JSDL specifications in terms of: 'OR' and 'NEGATION', currently there is only 'AND' NSF 08-571 Job Description   Agreed YES 2010-04-28
JD25   When the client wants to perform client initiated data-staging-in, the client MUST also specify in the Job Description via holdpoints that the Activity holds   Job Description, Activity Management / State Model   Dropped (Dupl. JM9.1) 2010-04-23


SM)  Requirements applying mainly to Activity Management / State Model (2010-03-26: 33 Req.)

Nb ID Description Source Areas Dependencies Status Date
SM1 119 The Execution service MUST support a common state model General Activity Management / State Model   Proposed 2010-03-18
SM2 120 The Execution service MUST expose one or several state where the Activity is finished (with success, error, cancellation, failure, ...) Etienne Urbah Activity Management / State Model   Proposed 2010-03-23
SM3 121 The Execution service MUST NOT separate 'Payload Success' (= 'Application Success') and 'Payload Error', which are both detected by the Payload itself, and where Payload and Activity logs are complete Etienne Urbah Activity Management / State Model SM2 Proposed 2010-03-18
SM3.1 122 The State Model for simple Activities MUST have a single state grouping 'Payload Success' and for 'Payload Error' Etienne Urbah Activity Management / State Model SM3 Proposed 2010-03-18
SM3.2   The single state grouping 'Payload Success' and for 'Payload Error' is named 'Finished with Success or Error' Etienne Urbah Activity Management / State Model SM3.1 Dropped (Premature, Specification of implementation) 2010-04-27
SM4 123 The Execution service MUST clearly separate the 2 following concepts :
• 'Payload Error', which is detected by the Payload itself, and should be investigated by Application specialists first
• 'Activity Abortion', which is caused / managed by the Execution Service, and require grid skills to be investigated
Etienne Urbah Activity Management / State Model SM2 Proposed 2010-03-18
SM4.1 124 The State Model for simple Activities MUST have separate states for 'Payload Termination' and for 'Activity Abortion' Etienne Urbah Activity Management / State Model SM4 Proposed 2010-03-18
SM4.2 125 Inside 'Activity Abortion', the Execution service MUST clearly separate the 2 following concepts :
• 'Activity Failure', which happens suddenly, is detected by the Execution Service itself, and where Activity logs can be partial or missing
• 'Activity Cancellation', which is triggered by the Client, cleanly processed by the Execution Service, and where Activity logs are complete
Etienne Urbah Activity Management / State Model SM4.1 Proposed 2010-03-18
SM4.3   The state for 'Activity Failure' is called 'Failed' Etienne Urbah Activity Management / State Model SM4.2 Dropped (Premature, Specification of implementation) 2010-04-27
SM4.4   The state for 'Activity Cancellation' is called 'Canceled' Etienne Urbah Activity Management / State Model SM4.2 Dropped (Premature, Specification of implementation) 2010-04-27
SM5 126 On Activity failure, the Execution service MAY, as many times as specified in the Job Description document, perform an automatic resubmission of the Activity by setting the Activity state to 'Submitted' General Activity Management / State Model JD5 Proposed 2010-03-18
SM6   The Execution Service MUST have an 'Output Data Retrieval' functionality :
When the Activity is finished (with success, error, cancellation, failure, ...), if the Execution Service keeps Data related to the Activity (for example an 'Output Sandbox'), then the Execution Service MUST allow the Client to request the retrieval of this Activity Data, and the Execution Service MUST send this Activity Data to the Client
Etienne Urbah Activity Management / State Model, Data Management SM2 Dropped (Duplicate) 2010-04-27
SM6.1   The Execution Service MUST have a 'Purge' functionality :
When the Activity is finished (with success, error, cancellation, failure, ...), the Execution service MUST be able, after the Client has retrieved Output Data of the Activity or after a long delay, to purge the Activity, that is to delete all Data related to the Activity (but the Execution service MUST NOT delete the Activity log).
Etienne Urbah Activity Management / State Model, Data Management SM6 Dropped (Duplicate) 2010-04-27
SM6.2   After the Execution Service has purged an Activity, the Activity log and Accounting MUST still be accessible (this MAY be achieved, for example, by separate 'Logging & Bookkeeping' and 'Accounting' services) Etienne Urbah Activity Management / State Model LB1, SM6.1, conflicts with SM6.3 Re-proposed 2010-04-28
SM6.3   After an Activity has been purged, the Execution Service SHOULD NOT keep any information on this Activity, and SHOULD answer to any request on this Activity with an error message   Activity Management / State Model SM6.1 Re-proposed 2010-04-28
SM6.4 127 Requirement to have a 'Purged' state, where 'Purge' refers to cleanup the whole existence of an Activity General Activity Management / State Model SM6 Proposed 2010-03-18
SM7 128 The Execution service MUST clearly separate the 2 following high level Activity states :
• 'Activity Submitted', where the Activity is (at least partially) validated, but storage / computing resources have perhaps NOT been allocated to the Activity yet
• 'Activity Processing', where the Activity has already been fully validated, and storage / computing resources are being allocated or have already been allocated
Etienne Urbah Activity Management / State Model   Proposed 2010-03-18
SM7.1   The state for 'Activity Submitted' is called 'Submitted' Etienne Urbah Activity Management / State Model SM7 Dropped (Premature, Specification of implementation) 2010-04-27
SM8 170 In order to permit the Client to perform direct manual processing inside submitted Activities (for example manual data staging), the State Model MUST contain
• one State permitting manual processing before Payload execution
• one State dedicated for Payload execution
• one State permitting manual processing after Payload execution
Etienne Urbah Activity Management / State Model   Re-proposed 2010-04-28
SM8.1   The state for 'Manual processing before Payload execution' is called 'Pre-processing' Etienne Urbah Activity Management / State Model SM8 Dropped (Premature, Specification of implementation) 2010-04-27
SM8.2   The state for 'Payload execution' is called 'Delegated' Etienne Urbah Activity Management / State Model SM8 Dropped (Premature, Specification of implementation) 2010-04-27
SM8.3   The state for 'Manual processing after Payload execution' is called 'Post-processing' Etienne Urbah Activity Management / State Model SM8 Dropped (Premature, Specification of implementation) 2010-04-27
SM9   The Execution service MUST set the Activity on 'Hold' at the Activity states specified inside the Job Description document Etienne Urbah Activity Management / State Model, Job Description JM9, JD5 Dropped (Duplicate) 2010-04-27
SM9.1 129 Following Activity states MUST permit 'Hold' :  'Pre-processing',  other active states,  'Post-processing' General Activity Management / State Model SM8, SM9 Proposed 2010-03-18
SM9.2   The 'Submitted' Activity state MUST also permits 'Hold' Etienne Urbah Activity Management / State Model SM9 Dropped (Premature, Specification of implementation) 2010-04-27
SM9.3 130 For each state permitting 'Hold', the 'Hold' state MUST be implemented as a substate General Activity Management / State Model SM9 Proposed 2010-03-18
SM10   The Execution service MUST allow the Client to request a transition between Activity states Etienne Urbah Activity Management / State Model   Dropped (Duplicate) 2010-04-27
SM10.1 131 The Execution service MUST perform the transition between Activity states requested by the Client ONLY if it makes sense.  Otherwise, the Execution service MUST return an error to the Client. General Activity Management / State Model SM10 Proposed 2010-03-18
SM11 132 By default, the Activity should not end up in a 'Hold' state. General Activity Management / State Model   Proposed 2010-03-19
SM12 133 State model should enable the monitoring of the data-staging process (i.e. more feedback to the end-user via substates or similiar methods) GIN Activity Management / State Model SM8 Proposed 2010-03-19
SM13 134 The Execution Service SHOULD have a hierachical state model consisting of 1-st level, 2nd-level, 3rd-level, ....states.  Lower level states always sub-states of their parent-state. Strawman Functionality Activity Management / State Model   Proposed 2010-03-20
SM13.1 135 First two levels are mandatory or a reasonable level. Draft Specification Activity Management / State Model SM13 Proposed 2010-03-20
SM14   Maybe a realization w/o hierarchical state models.   Activity Management / State Model   Dropped (Duplicate SM13) 2010-04-27


DM)  Requirements applying mainly to Data Management (2010-03-26: 17 Req.)

Nb ID Description Source Areas Dependencies Status Date
DM1   During Activity submission, the Execution Service MUST permit the Client to transfer small files from the Client's local storage to the Execution Service (for example the 'Input Sandbox').  Once the Execution Service has allocated an adequate Storage Resource to the Client's Activity, the Execution Service MUST transmit these small files to this Storage resource. Etienne Urbah Data Management, Activity Management   Dropped (Duplicate) 2010-04-27
DM2   The Execution Service MUST perform service-directed (was automatic) data staging in and out. Etienne Urbah Data Management, Activity Management JD4 Dropped (Duplicate) 2010-04-27
DM3 136 The Execution Service MUST have an 'Output Data Retrieval' functionality :
When the Activity is finished (with success, error, cancellation, failure, ...), if the Execution Service keeps Data related to the Activity (for example an 'Output Sandbox'), then the Execution Service MUST allow the Client to request the retrieval of this Activity Data, and the Execution Service MUST send this Activity Data to the Client
General Data Management, Activity Management / State Model   Proposed 2010-03-18
DM4   In order to apply the principle of 'Full Automation of Processing' and to prevent too much complexity inside the Activity State Model, specifications will NOT take into account 'Hold' points for Activities Etienne Urbah Data Management, Activity Management   Dropped (Dupl. JM8) 2010-03-18
DM5   In order to permit the Client to perform manual data staging, the Execution Service MAY manage 'Hold' points for Activities.  Fault if not supported. Etienne Urbah Data Management, Activity Management / State Model JD6 Dropped (Dupl. JM9) 2010-03-18
DM5.1 137 If the Execution Service manages 'Hold' points for Activities, it SHOULD be able to detect which computing resources (GLUE ExecutionEnvironments) run several Activities per core, and which run at most 1 Activity per core Etienne Urbah Data Management, Activity Management DM5 Proposed 2010-03-18
DM5.2 138 If the Execution Service has detected that the computing resource runs several Activities per core, then the Execution Service MAY set running Activities to a 'Hold' State (waiting for Client action) without under-utilizing available cores Etienne Urbah Data Management, Activity Management DM5 Proposed 2010-03-18
DM5.3 139 If the Execution Service has detected that the computing resource runs at most 1 Activity per core, then the Execution service SHOULD prevent running Activities to go to a 'Hold' State (waiting for Client action), in order to prevent under-utilizing available cores Etienne Urbah Data Management, Activity Management DM5 Proposed 2010-03-18
DM5.4 140 If the Execution Service is NOT able to detect if the computing resource runs several Activities per core or only 1, then the Execution service SHOULD prevent running Activities to go to a 'Hold' State (waiting for Client action), in order to prevent under-utilizing available cores Etienne Urbah Data Management, Activity Management DM5 Proposed 2010-03-18
DM6   In order to permit the Client to perform manual data staging, the Execution service MUST allow the Client to request information about Storage resources allocated to the Activity, and return the requested information (or an warning if Storage resources have NOT been allocated yet) Etienne Urbah Data Management, Activity Management   Dropped (Duplicate) 2010-04-27
DM6.1 171 One of the storage resources allocated to the Activity MUST be named 'Session Directory' Etienne Urbah Data Management, Activity Management DM5, DM6 Re-proposed 2010-04-28
DM6.2   If the Execution Service manages 'Hold' points for Activities, it MAY also be able to verify that manual data staging is really complete, and keep the Activity on 'Hold' otherwise Etienne Urbah Data Management, Activity Management DM5, DM6 Dropped (Premature, Detailed specification) 2010-04-27
DM7 141 Any client should be able to interact with the 'session directory' of any Activity created by the Execution Service GIN Data Management, Activity Management DM6, DM6.1 Proposed 2010-03-19
DM8 142 The Execution Service MUST support client-directed data staging in and out by exposing the locations of the Activity processing directory/session directory. GIN Data Management, Activity Management   Proposed 2010-03-19
DM9 143 Data Staging MUST support the following protocols (not defined in the document, added by Grimshaw to start discussion): HTTP(S), scp, gridftp, mailto, RNS/ByteIO GIN Data Management   Proposed 2010-03-18
DM10 144 Minimize the amount of calls required to interact with Execution Service (for example: in one WS: vector operations, service exposes the location where the client can stage any input data, service exposes the location where the client can fetch any output data.) GIN Data Management, Activity Management   Proposed 2010-03-19
DM11   In the context of data, any kind of data that has been marked as stage-in or any kind of temporarily generated data is not available for example: when activities enter the 'cancelled' state (or other final states?!).  Data that is marked as stage-out, on the other hand, MUST be available if present.   Data Management, Activity Management   Dropped (Premature, Detailed specification) 2010-04-27


 




The Open Grid Forum Contact Webmaster | Report a problem | GridForge Help
This is a static archive of the previous Open Grid Forum GridForge content management system saved from host forge.ogf.org file /sf/wiki/do/viewPage/projects.pgi-wg/wiki/RequirementsTable at Thu, 03 Nov 2022 00:04:34 GMT