Dr. Tony Doyle – A.Doyle@physics.gla.ac.uk

Dr. Steve Lloyd – S.L.Lloyd@qmw.ac.uk

1. Resource Allocation *

2. GridPP Financial Comparison with other Grid Projects *

3. International Grid Collaborations *

4. Grid Architecture *

5. Links with Industry *

6. APPENDIX *

1. Resource Allocation

1. Introduction

The programme for the UK Grid for Particle Physics consists of a series of interlocking "components". Components 1 to 4 are considered to be essential to meet the requirements of the GridPP programme.

Table 1 shows the PPARC funds invested as a function of component to CERN and the UK.

Table 1: Top-level financial breakdown.

Component	Description	CERN contribution from PPARC (£M)	Cost to PPARC (£M)	Total PPARC (£M)
1	Foundation	2.5	8.5	8.5
2	Production	1.0	4.1	12.7
3	Middleware	2.1	4.4	17.0
4	Exploitation	1.5	4.0	21.0
5	Value-added exploitation	2.9	4.9	25.9

Integrated over the first 4 components, the amount required for funding CERN activities is £7M.

The financial breakdown into various categories is given in Table 2 as a function of year.

Table 2: Financial breakdown by year for components 1-4.

Category	to Apr-02	Costs to Apr-03	to Apr-04	Total (£M)
Tier Centres Staff	0.30	0.81	1.35	2.46
Capital	1.06	0.87	1.03	2.96
Work Groups Staff	1.24	3.38	3.21	7.84
Capital	0.05	0.10	0.10	0.24
CERN Component Staff	0.80	2.63	2.23	5.67
Capital	0.38	0.47	0.56	1.42
Managers	0.08	0.16	0.16	0.41
Totals	3.91	8.43	8.64	20.98

 

Clearly the total of £3.91M to Apr-02 is not within the range of £2-2.5M currently discussed so a certain amount of re-profiling will be necessary. However, since these numbers come from a bottom-up assessment of the effort needed to produce specified deliverables to a given time scale it is not sensible to perform a detailed re-profiling until the total project budget, and therefore the project scope, is defined. All the numbers in this document clarify the model presented in April. Once the funding profile is known the Project Management Board will revise the planning to optimise the project in the light of both financial and technical developments in the interim.

The Staff Year (and corresponding financial) breakdown into WorkGroups is given in Table 3 as a function of year, separated into the funding already committed by PPARC to the EU DataGrid (EUDG) and the PPARC funding for GridPP with the DataGrid component excluded (UKPP). The mapping between the GridPP WorkGroups and the EU DataGrid WorkPackages was provided in Table 11 of the Proposal. For some EUDG WorkPackages the mapping is not unique, in which case it is assigned to one WorkGroup for ease of identification. For example, WP8 maps onto WG-H and WG-I and the total of 21 FTE of EUDG effort is assigned to WorkGroup H.

Table 3: Staff Years by WorkGroup as a function of year for components 1-4.

WorkGroup	Year-1		Year-2		Year-3		Total
	UKPP	EUDG	UKPP	EUDG	UKPP	EUDG	UKPP	EUDG
A	1.12	0.50	2.15	0.25	2.13	0.75	5.40	1.50
B	0.70	0.51	2.20	4.11	3.10	2.88	6.00	7.50
C	0.00	1.50	1.80	0.00	2.40	0.90	4.20	2.40
D	2.25	0.00	3.50	0.75	4.75	0.75	10.50	1.50
D	0.00	1.60	0.00	1.45	0.00	1.45	0.00	4.50
E	1.10	0.00	2.66	0.00	2.70	0.00	6.46	0.00
F	1.70	1.13	3.07	2.27	2.73	2.60	7.50	6.00
G	3.50	0.00	14.00	0.00	20.00	0.00	37.50	0.00
H	1.50	4.93	5.25	8.03	5.75	8.03	12.50	21.00
I	4.50	0.00	21.18	0.00	20.50	0.00	46.18	0.00
J	2.00	0.00	5.00	0.00	3.00	0.00	10.00	0.00
Totals (SY)	18.37	10.18	60.81	16.86	67.06	17.36	146.24	44.40
	28.55		77.67		84.43		190.64
Totals (£M)	0.99	0.55	3.28	0.91	3.62	0.94	7.89	2.40
	1.54		4.19		4.56		10.29

A further breakdown according to the deliverables described in the Proposal is given in Table 5 of the Appendix, incorporating CERN-based activities and external funding from the EU.

2. Funding Scenarios

We have been requested to plan for a programme in the range of £15-20M. This is at the point where elements of the Component 3 (Middleware) and Component 4 (Exploitation) need to be considered.

Component 3 enables the UK and CERN to make suitable contributions towards generating the middleware required for full Grid functionality, and in the process create generic software which will be used by other disciplines and in wider contexts. Component 4 is focused on providing the applications to deliver the science using the Grid infrastructure.

Components 1 to 4 must be funded, as an absolute minimum, in order for the GridPP programme to be realised



 

Figure 1: Breakdown of GridPP Resources for components 1-4 (£21M). The WorkGroups are: A Workload Management; B Information Services and Data Management; C Monitoring Services; D Fabric Management and Mass Storage; E Security Development; F Network Development; G Prototype Grid; H Software Support; I Experimental Objectives; J Dissemination; K CERN. The EU DataGrid components are denoted by an asterisk.

In the budget scenarios discussed below, we make the assumption that the total CERN component scales with the total but that the PPARC commitment to the EU DataGrid cannot be reduced. Furthermore, we apply a threshold below which we do not feel it sensible to arbitrarily scale elements of the project. Reductions beyond this threshold are achieved by reducing the scope of the project.

2. £20M Budget

In the £20M scenario, the CERN component is reduced by £0.4M to £6.7M where the savings would focus on hardware. A similar reduction is applied to the UK capital budget, which is reduced by £0.3M to £2.9M by reducing hardware costs by 5% each year. The remainder of the programme, excluding the EU DataGrid commitment, is scaled by 96.3% to achieve a further saving of £0.3M.

 



Figure 2: Breakdown of GridPP Resources for a £20M Budget. The EU DataGrid components (already funded) are displaced. The reductions and scaling factors discussed in the text for the non-EU DataGrid components are given.

4. £17M Budget

We have examined the impact of a £17M program and conclude that it is not possible to simply scale back the existing Proposal. The solution we would propose is to reduce, pro-rata, the CERN component to £6M and the UK capital spend to £2.45M giving a total saving of £1.85M. We would not scale back the commitments to the EU DataGrid, but the other elements of the programme except for the experiment objectives (WG-I) would be funded at an average level of 90%, saving a further £0.54M. This would require input from the annual review mechanisms that are envisaged within the programme. Once a level of funding is approved, the detailed work packages will be put out to tender by the GridPP project management and all groups will be invited to bid for this work. Bids will be fully peer reviewed and will be selected on the basis of competence, commitment and current involvement of the groups. To achieve the total programme cost of £17.3M, a further savings of £1.3M is made by cutting in half Component 4 (Experiment Objectives), which would be reduced from £2.5M to £1.2M. In practice, this is a false economy since this work would have to be funded by the experiments from other elements of the core programme.



 

Figure 3: Breakdown of GridPP Resources for a £17M Budget. The EU DataGrid components (already funded) are displaced. The reductions and scaling factors discussed in the text for the non-EU DataGrid components are given.

6. £15M Budget

Within the component model, it is impossible to achieve the programme described in the Proposal with £15M. The CERN component would have to be scaled back to £5M, saving an additional £1M. The other £1.2M saving would have to come from eliminating the experimental objectives (Component 4) completely from the scope of this project. The overall balance of the GridPP project would be radically distorted. We stress that the concurrent development of the applications is essential to provide feedback during the development of the Grid.



 

Figure 4: Breakdown of GridPP Resources for a £15M Budget. The EU DataGrid components (already funded) are displaced. The reductions and scaling factors discussed in the text for the non-EU DataGrid components are given.

 

 

 

 

 

3. Experiment Objectives

The computing requirements of the LHC were described in the Proposal: development of the storage, management, simulation, reconstruction, distribution and analysis of the data of the four LHC experiments (ALICE, ATLAS, CMS and LHCb) constitutes an unprecedented challenge to the High Energy Physics (HEP) and Information Technology (IT) communities. The design of the software, both experimental and generic, to exploit Grid computing needs to start now in order to evaluate fully its performance and assess whether it will meet future demands. In parallel to addressing these demands, it is also crucial that the experiments generate large amounts of Monte Carlo simulated data in order to understand and optimise the design of the detectors to maximise the physics return in 2006. Both these needs call for substantial investment in computing infrastructure, both in the area of computing hardware and software development.

Less emphasis was placed on the integration of the current US experiments in the Proposal: the concurrent development of applications driven by real data, with immediate requirements of Grid technologies is also important. Examples of current developments of this aspect of the programme are described below.

The D0 experiment has developed a sophisticated data-handling tool known as SAM. The UK is implementing job submission and resource management aspects into SAM; this involves incorporating a Grid enabled submission mechanism. The project is being carried out in the context of the PPDG project and in close collaboration with the Condor team, based at Wisconsin University. Other UK personnel on D0 are heavily involved with Grid interfaces to mass storage elements within PPDG. Contacts are already developing between this work and the UK led DataGrid work package in this area.

In the UK, BaBar will develop a Tier-A processing centre that will meet a significant fraction of the whole collaboration’s computing needs. Working directly with PPDG, it is intended to implement a version of the storage request broker (developed at the San Diego Supercomputer centre) that is Grid enabled at this centre. In addition, a UK distributed Monte Carlo production system is currently being commissioned based on 9 PC farms. It is planned that this production system will be integrated into the Grid environment. BaBar plan to use the PPDG/Globus tools for bulk data transfer and the UK currently co-ordinates the BaBar developments in this area.

CDF’s computing model was developed based on their Run-I experience and the assumption that network connectivity between Europe and the States would continue to be inadequate. The model was developed very early in the CDF upgrade plans and, as a consequence, a major investment in their software environment, outside of a distributed framework, was made. Even so, the UK is driving forward the CDF Grid effort in collaboration with US institutes. Early work has included making the CDF metadata available via LDAP so that each collaborating site can publish a list of available resources and datasets. A feasibility study has also been undertaken which shows that the existing Disk Inventory Manager can be made Grid aware in a simple way. Support through GridPP for the CDF developments would allow these efforts to meet their full potential and spawn possible future funding in the US for CDF Grid effort.

The UK is playing a leading role in developing the middleware of the DataGrid. It is equally important that the UK take advantage of their prominent positions with the experimental collaborations in order to ensure that GridPP is taking a leading role in stress testing and utilising the developing tools. Not only should the UK build on its leadership within the LHC experiments, in the context of the DataGrid project, but should also exploit its prominent position within the US collider experiments in the area of Grid computing. This would enable important links to be forged between the EU DataGrid project, through GridPP, and the GriPhyN and PPDG projects in the States; thus fostering a closer co-operation.

2. GridPP Financial Comparison with other Grid Projects

It is difficult to compare the Grid projects of other countries because of different policies on funding of staff and different profiles of spend. The following information has been determined from a meeting of the HEP-CCC in Bologna (June 15), where GridPP was represented, and focuses on the investment in Tier-1 Regional Centres and associated Tier-2 Centres in various countries. These show the UK plans as being similar to other countries with the most e-Science aware one, Italy, being significantly more aggressive.

1. US

1. CMS

The US plans a Tier-1 and 5 Tier-2 centres for CMS. A prototype will be built during 2000-04, with full deployment during 2005-7 (30:30:40 funding profile)

Staff estimates for the Tier-1 centre are 14 FTE by 2003, reaching 35 FTE in 2007.

Total Costs for Tier-1 and Tier-2 for 2001-2004 are $M(3.5, 4, 5.5, 8.5).

The integrated costs up to 2006 are $54.7M for hardware and software support for CMS. In addition, GriPhyN and PPDG (described in section 3) will also contribute to US LHC work. The currently approved budgets are $12.5M for GriPhyN1, $25M for GriPhyN2.

2. ATLAS

ATLAS plans a similar structure of Tier-1 and 5 Tier-2 centres for the same cost as CMS. We did not have access to detailed costs but estimates of the size is the same as CMS for CPU, less for disk and more for tape.

2. France

France plans a Tier-1 RC for all 4 LHC experiments at CC-IN2P3 in Lyon and an LHC prototype starting now. The size is similar to a prototype Tier-1 defined in the CERN Hoffmann report. Estimates but no details of manpower have been determined for IN2P3.

2M€/year for a National Grid has been assigned now by ministerial action. CC-IN2P3 are also working on a BaBar Tier-A centre and Grid work with BaBar. The Tier-1/A work will be a major part of CC-IN2P3 work in future years. Their annual budget is 8M€ and 40FTE.

3. Germany

Karlsruhe run a large computer centre for a variety of disciplines (50FTE and 11M€ annual budget). The Tier-1 RC for Germany will be sited there. It will make use of existing infrastructure with 8 additional staff and 12M€ for hardware assigned up to 2006. This is for Tier-1 only and provides no experimental support.

4. Italy

Italy plans a Tier-1 RC and a prototype starting now in CNAF, Bologna. Hardware plans are more aggressive than UK. 15.9M€ is allocated during 2001-3 for Tier-1 hardware and outsourced staff. Additional staff will be funded for Tier-1 support (7; 21; 25 FTE for each year). They also plan 10 Tier-2 centres at 1M€/year. An initial 24M€ budget has already been allocated.

3. International Grid Collaborations

1. General/Summary

• GridPP has formally engaged with the major grid projects in the US: GriPhyN, PPDG and lately iVDGL: we have agreed at high level to work with these projects, we have identified areas of collaboration already in progress and natural areas of collaboration to develop.
• In almost all cases the people involved in this work on both sides of Atlantic are already well known to each other, since the work covers an area of worldwide expertise which has already been developed.
• These engagements overlap with, and are consistent with the relation between these US projects and the EU DataGrid: the people involved are already working in a global environment.
• Two members of GridPP participate and enact within the DataGrid-Globus coordination group.
• Contact has been made with the Condor team, maintained via direct contact, and described below.

1. GriPhyN

1. PPDG

1. iVDGL

1. InterGrid Co-ordination Group

"International HENP Grid Coordination & Joint Development Framework"

Strategic co-ordination of Particle Physics Grid projects has been the subject of meetings between representatives from EU-DataGrid, GriPhyN, PPDG and Asian activities. So far these have taken place in Amsterdam (March 2001) and Rome (June 2001). The UK has been represented by GridPP and by PPARC. It is foreseen that in future PPARC will be represented by the Director e-Science. The group also has amongst its members the Principal Investigators of the Globus and Condor projects. The next meeting is foreseen to be in October.

1. Structure

After much discussion it has been agreed to operate a 2-tier structure. The first part is a Management Board responsible for overall co-ordination. The second part is a Technical Board charged with ensuring that all technical issues are covered either by launching task forces of domain experts, by ensuring joint activities of existing projects or by requesting new (joint) projects (the latter passed to the Management Board for approval and initiation). The exact terms of reference of these two bodies are currently being developed.

An additional tier of common projects management has also been discussed, but deemed to be of limited benefit, since such bodies would have no direct resources but would in any case rely on the funding available through core participants.

The existing body is to become the Management Board and nominations are currently sought for the Technical Board (6 US, 6 Europe, 2 Asia) in order that it may start its work before the end of the summer. GridPP has proposed one member to sit on this technical board.

2. Aims

The overall aim of the co-ordination is primarily to promote common activities with a view to ensuring seamless interfaces for applications in the various countries and to avoid (or at least minimising) duplication. To date it has been decided to promote joint testbeds, grid architecture co-ordination and to work for a common Open Source license structure. The DataTAG project and its association with iVDGL is in part a consequence of this co-ordination activity.

3. Open Source License

Work has also progressed on software licensing issues with an outline for a Consortium for Open Grid Software (COGS) having been discussed. Associated Open Source license text is already available and has been distributed to participating institutes of GridPP for comment and feedback. It is foreseen that developments of the EU DataGrid will be made available under this license.

2. Common Testbeds

As part of the first testbed release of the EU DataGrid project it is planned to have the direct participation of up to 4 sites in the US in the testbed and is anticipated to include at least the prototype Tier-1 centres. The GridPP prototype activities will this be fully linked with this.

4. Future

1. Grid Architecture

Our overall aim is to make sure that the Grid is a reality for PP in the UK: we feel that the best way to achieve this is by making individual contributions to architectures via our links to the Global Grid Forum (GGF), EU DataGrid, PPDG and GriPhyN. GridPP does not plan to have its own architecture body.

We support the GGF in its attempts to promote best practices and to establish standards and will take particular note of the activities of the recently formed architecture group within the GGF. We expect to see standards emerge over the next few years and we are especially concerned that architectures should be capable of evolution.

We have a strong link to the Architecture Task Force (ATF) of the EU DataGrid, with three UK members on that body, one of whom is currently the chair.

We also have direct links to the National e-Science Centre and five of the eight Regional e-Science Centres, which will provide a forum for architecture development in the future. The National Centre Director and the Chief Software Architect who supports the development of the research strategy will have an important role in ensuring the evolutionary capabilities of the architecture.

The GGF Grid Protocol Architecture Working Group draft charter is noted below. Various members of GridPP will be attending the GGF meeting in Washington and contributing to discussions in the Grid Protocol Architecture Working Group. The GridPP project leader is a member of this Working Group and will provide a link to ensure that the architecture adopted by GridPP is fully consistent with that developed in the Grid Forum.

1. GGF Grid Protocol Architecture Working Group

1. Draft Charter

The role of the Grid Protocol Architecture Working Group is to provide a conceptual framework for discussing the interrelationships, completeness, and minimality of the protocol approach to Grid services that is coming out of GF.

2. Proposed Goals

1. Define an architecture for the protocols, services, and API model of Grids.
2. Draft an architecture document which will identify Grid functions and services, and their relationship to applications, resources, and the other services. The document will also attempt to identify a minimally complete set of functions and services.
3. Examine the work of the other WorkGroups in the context of this architecture and comment on both minimality and completeness of the overall GF work.

1. Links with Industry

1. Introduction

The PP community intends to build upon its existing links with industry. We intend to pursue these links at several levels. We will develop high level relationships through our Dissemination Board and grass roots contacts through our Regional Centres and Institutes. Industrial partners are already working with PP groups in JREI and JIF projects, in particular in the provision of compute-engines and mass-storage facilities. These facilities are part of the core infrastructure for the national GridPP, and the industrial partners have joined with us to gain from our experiences. The current investment in hardware and associated industrial partnership is shown in Table 4.

Table 4: External Funds (additional to PPARC Grants and central facilities) providing computing equipment and the basis for Industrial Partnership

Collaboration	Cost (£M)	Industrial Partner	Funding
BaBar (Birmingham, Bristol, Brunel, Edinburgh, Imperial, Liverpool, Manchester, QMUL, RAL, RHUL)	0.8 1.0	Sun	JREI JIF
MAP (Liverpool )	0.3	ITS	JREI
ScotGrid (Edinburgh, Glasgow)	0.8	IBM	JREI
D0 (Lancaster)	0.4 0.1	WS-UK	JREI Univ.
Dark Matter (Sheffield)	0.03	Tbd	JIF
CDF/Minos (Glasgow, Liverpool, Oxford, UCL)	1.7	IBM	JIF
CMS (Imperial)	0.15	Tbd	JREI
ALICE (Birmingham)	0.15	Tbd	JREI
Total Cost	5.4

Our aims with respect to industrial partnership are to: help us develop the Grid by supplying hardware and software; use the Grid for themselves for collaborative engineering, massive simulations and federating their own worldwide databases; sell or develop the Grid for other early adopters. This links to UK developments via the National e-Science Centre: we are keen to ensure that each of the GridPP Regional Centres benefits from a co-ordinated approach to industrial partnership, a task which will be taken up by the Dissemination Board.

2. IBM-UK

As part of its growing commitment to the Grid, IBM has dedicated resources through various collaborations with Institutes throughout the UK. The process of identifying appropriate Grid tools from IBM’s existing strategic development areas in e-Utilities, e-business (project eLiza), e-markets and Intelligent Infrastructure with respect to the Grid architecture has started. Figure 5 illustrates this relation: there is a clear mapping of the Collective, Resource, Connectivity and Fabric layers to the Middleware WorkGroups in GridPP. Various Institutes are currently identifying areas of common interest.



Figure 5: Layers within the Grid mapping to IBM technologies. Each one of the layers will have its own Application Programming Interfaces (APIs) and Software Development Kits (SDKs).

An early example of a collaborative project with IBM-UK is the development of a Producer/Consumer model for the Grid Monitoring Architecture (GMA), interfaced to an SQL service. This relational model enables both live and archived monitoring data to be efficiently queried with a common interface. The GGF performance architecture does not currently specify the protocol between the Producer and Consumer: the relational model has been developed consistent with the GMA and was recently demonstrated at the Oxford EU DataGrid Meeting.

We envisage a close relationship in the development of the Fabric, Connectivity and Resource layers with vendors for the UK Tier centres. We also envisage relationships with database providers such as Objectivity, Oracle, MySQL and PostgreSQL from the Application and Collective layers. The PP community will take the best advantage of the opportunities of industrial liaison to enhance GridPP, and in turn will pro-actively seek to aid the transfer of technologies, rather than being a passive application driver.

2. APPENDIX

1. Deliverable Requirements