Application Hosting Services for Research Community on Multiple Grid Environments

Transcription

1 Hosting Services for Research Community on Multiple Grid Environments Hiroyuki Kanazawa, Naoki Onishi, Yuri Mizusawa, Takahiro Tsunekawa, Hitohide Usami Hosting Services for Research Community on Multiple Grid Environments 1,2 Hiroyuki Kanazawa, * 3 Naoki Onishi, * 4 Yuri Mizusawa, * 5 Takahiro Tsunekawa *6,*7 Corresponding author Hitohide Usami * 1 Technical Computing Solution Unit, Fujitsu Limited, Nakase 1-9-3, Mihama-ku, Chiba-city, Chiba , Japan, kanazawa.h@jp.fujitsu.com 2 Graduate School of Natural Science & Technology, Kanazawa University, Kakuma-machi, Kanazawa, Ishikawa, , Japan *3 Technical Computing Solution Unit, Fujitsu Limited, Nakase 1-9-3, Mihama-ku, Chiba-city, Chiba , Japan, onishi.naoki@jp.fujitsu.com *4 Technical Computing Solution Unit, Fujitsu Limited, Nakase 1-9-3, Mihama-ku, Chiba-city, Chiba , Japan, yuri@jp.fujitsu.com *5 Technical Computing Solution Unit, Fujitsu Limited, Nakase 1-9-3, Mihama-ku, Chiba-city, Chiba , Japan, tsunekawa.takah@jp.fujitsu.com *6 Research Institute, Tamagawa University, Machida-city, Tokyo , Japan, usami@nii.ac.jp *7 Center for Grid Research and Development, National Institute of Informatics, Hitotsubashi, Chiyoda-ku, Tokyo , Japan doi: /jcit.vol5.issue4.16 Abstract This paper describes the application hosting services (AHS) that is one of software component of multiple grid environments middleware developed by RENKEI (REsources linkage for E-scIence) project. The purpose of this middleware realizing the federate/share resources (computers, storages, databases and applications) distributed among multiple grid environments, such as research laboratories, national computer centers and international grids etc, AHS provides repository of application information to the research community created by grid virtual organization (VO). Community members can share the applications and their execution environment on the multiple grid environments. Keywords: Hosting Services, Grid, PSE, Research Community and Grid virtual organization (VO). 1. Introduction The National Research Grid Initiative (NAREGI) project conducted by the Ministry of Education, Culture, Sports, Science and Technology, Japan (MEXT) finished in FY2007 [1].The NAREGI grid middleware Ver.1.0 (RPM packaged version), which was the result of the research and development, was delivered on May 9, Several Japanese national supercomputer centers has installed and developed management tools for the grid operation. The National Institute of Informatics (NII) established a support organization called the Grid Operation Center (GOC) for the smooth deployment of the NAREGI grid middleware to users. On the other hand, a new project called RENKEI (REsources linkage for E-scIence) has been started [2]. "RENKEI" is Japanese word meaning "federation". The RENKEI is a 4-year project that will continue until the end of FY2011 and its main activity is the extension of the NAREGI grid middleware for adding the laboratory level resources on the community, for example, small cluster machines, experimental equipments, and so on in the laboratory [3]. We have developed NAREGI-PSE (Problem Solving Environments) on the NAREGI project [4]. This time, we are developing software in view of new concept the application hosting called AHS. This system is based on the NAREGI-PSE and inherits the community model, including user management in VO. 152

2 Journal of Convergence Information Technology Volume 5, Number 4, June RENKEI project RENKEI is a new project designed for a software environment where researchers can create and operate research community easily on the multiple grid environments. Specifically we are conducting research and development of grid middleware to utilize federated computing power, files, DBs etc, on different grid environments, from small scale laboratory level grid environments (LLS: Laboratory Level System), to large scale supercomputer center level grid environment (NIS: National Infrastructure System) and other foreign countries grid environments. computation intensive applications users database users application developers (1) computing resource federation, and application hosting (3) DB federation, and federation with ID management systems (4) API for multiple grid middleware DB laboratory (LLS) grid middleware (e.g. NAREGI) computer centers (NIS) DB DB international inter operation DB grid middleware (5) evaluation and collaboration with users computer users (2) file sharing, and file catalogue federation computation/data intensive application users Figure 1. RENKEI project overview and relationship of each groups RENKEI project consist of 5 R&D groups as follows. 1) Computing resource federation, and application hosting 2) File sharing, and file catalogue federation 3) DB federation and federation with ID management systems 4) API for multiple grid middleware 5) Evaluation and collaboration with users Figure 1. shows RENKEI project overview and relationship of each group Computing resource federation, and application hosting Community members can develop and run shared application programs distributed on various grid resources without knowing the details of any particular underlying grid middleware. This group developed three software components. 1) Hosting Service (AHS) 153

3 Hosting Services for Research Community on Multiple Grid Environments Hiroyuki Kanazawa, Naoki Onishi, Yuri Mizusawa, Takahiro Tsunekawa, Hitohide Usami AHS provides a software environment for deploying and sharing the scientific applications on NAREGI grid resources and on laboratory resources. developers use the AHS to register and deploy applications and to share them within research community members. 2) Workflow Tool (WFT) WFT provides end users with a single seamless interface to manipulate workflow jobs. They use the WFT to discover appropriate resources, to specify workflow jobs consisting of applications deployed by using the AHS on various grid resources, to launch jobs, to monitor running jobs, to terminate jobs and to retrieve files to client machine. 3) Interoperation system Interoperable function on NAREGI provides interoperation with EGEE(gLite) by developing jobsubmission and exchanging resource information through OGF(Open Grid Forum) standards (e.g. BES). End users can launch jobs on glite resources using the WFT. Overseas glite grid end users also can run File sharing, and file catalogue federation User can easily access to files on multiple grid environments without knowing the difference of each files specification. This group developed two software components. 1) File sharing system High performances file access from LLS/NIS with automatic replica file allocation and fault tolerance with automatic file replication. 2) File catalogue system Development of the file catalogue system with RNS (Resource Namespace Service) on OGF standard and federation of file catalogues running on different grid middleware (international interoperation) 2.3. DB federation and federation with ID management system User can easily query data on multiple DBs in multiple grid environments for research communities. This group developed two software components. 1) DB federation system Unified and easy access to multiple DB with different implementation, e.g. RDB, XML, Web, RDF, search engine, etc. on grid environments. 2) ID management system Generation of grid authority information federating with various ID management system API for multiple grid middleware User can easily develop application scripts running on multiple grid middleware. This group developed API to develop/run applications on multiple grid middleware with OGF standards, e.g. SAGA and RNS. 154

4 2.5. Evaluation and collaboration with users Journal of Convergence Information Technology Volume 5, Number 4, June 2010 Evaluation and collaboration with users are very important for the development of the practical usable middleware. This group activates two works. 1) Evaluation Building test beds with LLS/NIS, and evaluation of middleware technologies developed in the RENKEI project and feedback to the development teams. 2) Collaboration Collecting needs from application users for escience infrastructure. 3. Hosting Services (AHS) 3.1. Framework of AHS The purpose of AHS is to support the scientists works for wide range and daily researches (for example, experimental data processing, data analysis, modeling and large scale computer simulations, etc.). Virtual Organizations VO2 Groups Developers VO1 Users VO3 A Univ. B Univ. DUniv. C Univ. NIS Grid A Lab. B Lab. C Lab. D Lab. LLS Grid Joint Research Community Figure 2. Framework of Hosting Service Collaboration with research community members using the hierarchical large-scale grid environments effectively on the distributed computer systems from a high-end supercomputer system to a laboratory-level small computer system without detailed knowledge of the distributed computers configurations and operation method due to the difference of multiple grid environments. In a research community, application users may only know the application specifications such as the application name or application keywords for selecting the applications. However, application developers have relatively detailed knowledge of the applications such as input files or machine requirements. For effective collaboration, research community members should share such application knowledge within the community. Such application knowledge should fit grid environments. AHS mainly provide repository function of application files and information for research community using grid environments. (Figure 2.) 155

5 Hosting Services for Research Community on Multiple Grid Environments Hiroyuki Kanazawa, Naoki Onishi, Yuri Mizusawa, Takahiro Tsunekawa, Hitohide Usami 3.2. Collaboration in the research community For the demand in the collaboration of leading-edge research field, it is required to have a tool deploying an application including the repository to share necessary applications immediately in a grid environment Developing As for the application, development is pushed forward only with an individual or few developers till some quality is secured. It is necessary to deploy the application that worked in small environment easily in large-scale environment for the development of the application on the large-scale grid environment. Development can feed back a result of the execution in the large-scale environment that is really used by it quickly, then it can be developed efficiently. In addition, it helps promotion of efficiency to cover up complexity of the grid environment when grid environment is applied to. In addition, it is requested that an application does not leak out to the outsider carelessly at the time of the development in particular Sharing applications in the research community The developed application is provided to users or user groups who are collaborating with the developers closely. Because the application users have relatively less knowledge about the application than developers, it is necessary to cover up that information at the time of the use of the application. And also, detailed information about the grid environment should be covered up for application users. In addition, it is common that the application is updated frequently, and that the users expect the updated application be delivered immediately. Because an application is deployed on a large number of computers in the grid environment, the importance of this function increases Executing s on Grid At the run time of application, it is requested deploying an application on a large number of computers of the grid environment and executing it surely. In the deployment, it is necessary to select an appropriate computer resource satisfying the requirement of executing the application. Also, appropriate scheduling such as the selection of the computer with the lowest load is requested. Users build workflows for applications deployed on grid environment, and execute the workflows Architecture AHS basic architecture is constructed as a set of Web/Grid service infrastructure on top of WSRF (Web Services Resource Framework) specifications [5]. WSRF is a messaging model and provides the ability to model stateful resources in a framework of Web services. AHS operates cooperatively with the workflow tool and access to NAREGI-SS (Super Scheduler) for the NIS or GridSAM for the LLS as shown in Figure 3. AHS supports a single-sign-on system for both NIS and LLS grid environments using the NAREGI CA system. The user designs a WF research process by using the NIS and LLS execute icons on the WF GUI (another component of AHS) and the information for these applications is imported from the AHS application pool. Then, the WF module submits the NIS icon parts to the NAREGI-SS for the NIS resources and the LLS parts to the GridSAM for the LLS resources. Users easily change the execution resources only by replacing the execution icon on the WFT GUI. Scientific researchers usually conduct algorithm research or experiments using laboratory resources (LLS), but sometimes they need to perform large scale simulations or data analysis on supercomputer using NIS resources. In this manner, WFT and AHS are very useful for realizing the Everyday researchers doing everyday research. 156

6 Journal of Convergence Information Technology Volume 5, Number 4, June 2010 The NAREGI-SS selects the appropriate computers from the NIS resources on which the application is deployed. The GridSAM manages resources according to the assigned local schedulers. AHS communicates directly with the users to support them. Without the AHS, it is difficult to run the users complex research processes (for example, experiments -> data analysis (LLS site) -> modeling - > simulations (NIS site) -> next experiments) on geographically distributed computer environments. CLIENT (End Users) Portal server AHS Command Line Interface Portal Register & Delete Applicati Retrieval WFT on Pool Compile Deployment GridSAM LLS Resources Data share / DB share NAREGI SS/IS NIS Resources Figure 3. The architecture of Hosting Service Virtual Organization A virtual organization (VO) is a dynamic collection of resources and users unified by a common goal and potentially spanning multiple administrative domains [6]. In NAREGI middleware, VO is built based on VOMS architecture [7]. VOMS defines groups and roles hierarchically shown as Figure 4. Combinations of these names serve as attributes for users. Subheadings should be in a bold font and in lower case with initial capitals. There should be no left margin for them as well. There should be a blank line before and after them. The top of this paragraph illustrates a sub-subheading. AHS VO is designed completely same as NAREGI. Group_b Group_c Group_d Group_a Group_A Group_B Virtual_Organization Figure 4. Conceptual image of Virtual Organization (VO) 157

7 Hosting Services for Research Community on Multiple Grid Environments Hiroyuki Kanazawa, Naoki Onishi, Yuri Mizusawa, Takahiro Tsunekawa, Hitohide Usami pool Repository of applications, called application pool, of AHS is implemented by using the Contents Service (ACS) [8]. Contents Service (ACS) is a set of requirements and specifications that is being worked on by ACS-WG (Working Group) [9] within the OGF [10], which is an international standards and community body focusing of Grid technologies. The ACS Specification Documents defines a standardized way to manage and handle a grid application as a deployable logical unit so as to maintain their consistency, reduce management overhead, and enable automation throughout the lifetime of an application. Archive Producer CRUD history contents ARI Repository ARI subscribe notify contents Consumer Figure 5. Conceptual image of Contents Services (ACS) ACS specifies a repository for grid applications and its interface, Repository Interface (ARI), as shown in Figure 5. producer create Archive (AA) instances accompanied with the meta-information describing the contents. The AA is a logical bundle of files that is used for provisioning and executing a task in a Grid system. Such information may include, but is not limited to, the application executable code, configuration data necessary for initial deployment of the application and the deployment descriptor documents. Developer Create, Read, Update, Delete Contents AHS Pool = Information + ACS ( Repository) Repository AA Descriptor JSDL Contents Consumer Archive (AA) Figure 6. pool of Hosting Services and ACS An AA consists of an Archive Descriptor (AAD) and zero or more Contents. The ACS repository will parse and make use of the AAD. The AAD may contain information such as identification, access constraint policy, and information associated with the 158

8 Journal of Convergence Information Technology Volume 5, Number 4, June 2010 structure of the AA contents. In order to realize Contents retrieval, ACS repository needs to know the meta-information of the Content files, such as the names and types of the files. They should be provided by the Provider and be included in the AAD. These structures are shown in Figure Compiling source code and deploying executable on grids The Compilation Service provides function of remote compilation on the NIS and the LLS resources, according to the following procedure. The Deployment Service provides the deployment function. Figure 7 shows the application developer compile source code and deploy executable on grids. Developer User Registration of program A and program B. AHS Use of program B. proga progb A Deploy B Compile Compile Deploy A A B B #1 #2 #3 NIS Resource NIS Resource LLS Resource Figure 7. Compilation and Deployment On the other hand, an application will be required by other application. For example, an application using the function which is provided by other application may link library files when it is compiled or executed. In such situation, these related applications must be deployed on same computer resources. Therefore, application developer can register the applications required when the target applications are compiled or executed Access Control The administration of applications by AHS is different between applications in the application repository and applications in the disk of computer resources after deployment. In the application repository, access permission of applications can be controlled by AHS. On the other hand, on disks of computing resources, access permission of applications can be also controlled using AHS. But application files are in local file systems on each computer resources, then all user processes may be completely prohibited to access application files. It is supposed that user's home directory is usually safe, but files in the home directory cannot be accessed at all from other users who belong to the same VO. Providers of computing resources can place directory for sharing applications on disks of the computing resources. AHS will deploy application in that sharing directory if exists, otherwise in user's home directory. 159

9 Hosting Services for Research Community on Multiple Grid Environments Hiroyuki Kanazawa, Naoki Onishi, Yuri Mizusawa, Takahiro Tsunekawa, Hitohide Usami AHS Proxy Certificate (VO=VO1, Role=user) Proxy Certificate (VO=VO1, Role=admin) user User Interface Repository (VO) (personal) VO1 administrator Deployment Service Computing Resource (NIS) Computing Resource (LLS) VO1 $HOME $HOME (VO) (personal) (VO) (personal) Figure 8. Access Control on Repository and Computing Resources Execution The super scheduler of NAREGI (NAREGI-SS) is a scheduling system for large-scale control and management of a wide variety of NIS resources shared by different organizations in the grid environment [11]. The system will be aimed primarily at identifying resources that can meet requests from batch job users and allocating these resources to specific jobs. The GridSAM is a lightweight grid middleware for managing LLS resources. The grid workflow is a visual tool for seamlessly preparing and submitting distributed jobs running on NIS/LLS resources. It handles programs and data explicitly, and is independent of multiple grid middleware. Complex workflow descriptions such as loops and conditional branches are supported for science applications. Graphically described workflow jobs are converted to an enhanced workflow language based on WFML (WorkFlow Modeling Language), which is a common interface with other systems such as the AHS. Scientific researchers usually conducts small scale simulations or test executions using the laboratory level resources (LLS), but sometimes they require large scale simulations or data analysis on supercomputer using NIS resources. Users easily change the execution resources only by replacing the execution icon on the WFT GUI as shown in Figure 9. Large scale simulation on the NIS NIS icon Files (inputs) Execution Preprocessing Postprocessing File (results) Change the execution icon Chang the execution icon only LLS icon prg2 Files (inputs) Execution Preprocessing Postprocessing File (results) Small scale or test simulation on the LLS Figure 9. Grid Workflow 160

10 Journal of Convergence Information Technology Volume 5, Number 4, June Replication of Archives In large-scale science grid environment, many computer resources are widely-distributed geographically. In that case, the time of application deployment from a single application repository becomes long as the number of resources increases. The Replica, that is a part of AHS system, has Repository storing Archives which are duplicated from Archives in master Repository. The applications registered with the AHS are replicated in all other Replica s. Each Replica takes charge of some computer resources, which may be geographically near from that Replica. When an application is deployed to computer resources using AHS, files of the application are transmitted from nearer Replica to each computer resource. AHS User Interface Information Registration Deployment Repository (ACS) Replication Replica Repository (ACS) Deployment Service Deployment Service Computing Resources Computing Resources Replica Repository (ACS) Deployment Service Computing Resources Figure 10. Replication of Archives 4. Performance and evaluation AHS is currently being developed, and building a prototype system has just been finished. AHS has been installed in NAREGI facility which includes heterogeneous computing resource of grid test bed. We measured the performance of AHS primary functions using this NAREGI test bed Performance We had measured the performance (command execution time) of primary functions of ACS under following conditions. 1) Primary functions of ACS -Registration(src): (source code) register to the src file. -Registration(bin): (binary code) register to the bin file. -Compilation: Source code compile to object (not include compile time). -Deployment: deploy to target machine. 2) Hardware (PRIMERGY RX200) - Xeon 3.06GHz - 1GB memory 3) Software 161

11 Hosting Services for Research Community on Multiple Grid Environments Hiroyuki Kanazawa, Naoki Onishi, Yuri Mizusawa, Takahiro Tsunekawa, Hitohide Usami - RedHat9 - PostgreSQL ) The number of Login Users We assume the number of community members and login users as follows and the number of active login users is shown by % to the number of community members. -Large scale community : 500, Active login user (8%) : 40 -Typical community : 200, Active login user (10%) : 20 -Small scale community : 50, Active login user (20%) : 10 5) The number of s We assume the numbers of shared applications in application pool are 1, 20, 40, 80 and ) Target performance Response times of primary command execution are in 30 seconds at the typical use case (20 active users and 20 shared applications) The performance results are shown in Table 1. Table 1. Performance of Hosting Services Number of Users Number of s Registration (src) sec Registration (bin) sec Compilation sec Deployment sec Evaluation The performance is low under large number of active users (40) and applications (160), but for typical case of active users (20) and applications (20), the target performance as shown in Table 1 is achieved. The machine used for the measurement is old with small memory size. Taking into consideration the fact that the recent machine performance is rapidly increasing and large sized memory can be easily obtained, these performances will improve ten times faster than the measured results in a near future. Finally, we think these performances will be achieved somewhere in the region of use in actual community operations. 5. Conclusion AHS (successor version of NAREGI-PSE) is currently being developed, and building a prototype system has been just finished. NAREGI-PSE was already developed by spring, 2008 and released as one part of the NAREGI middleware Ver.1.0 treat VO as a flat group, without the hierarchical group structure. However, among users belonging to each VO, an application can be shared. As a part of the release of NAREGI middleware Ver.1.0, NAREGI-PSE works in large scale test bed including a heterogeneous computing resource in all over Japan. With the large-scale test bed building with ten universities in Japan, evaluation of collaboration by sharing applications in flat VO will be performed. The full-scale collaboration is to be evaluated after the implementation of a hierarchical group is completed. AHS offers a repository to share applications among researchers and a user group which conforms to a VO, and also offers a collaboration function to share applications. The Workflows are built with 162

12 Journal of Convergence Information Technology Volume 5, Number 4, June 2010 the Workflow Tool, and the applications for scientific research are executed via the super scheduler of NAREGI and LLS grid system. The expansion for collaboration of AHS and WFT is under development now. In the next spring, it will be released with other components of RENKEI prototyped version. By the release, large-scale test bed of NAREGI is used, and it goes to evaluation phase in the collaboration in science grid community. The scalability in the large-scale resource by the replica repository is going to be evaluated. 6. Acknowledgements This works was partly supported by a grant from the Ministry of Education, Sports, Culture, Science and Technology (MEXT) of Japan through the RENKEI (REsources linkage for E-scIence) project. We would like to express special thanks to Prof. Kenichi Miura (NII), Prof. Kent Aida (NII), Prof. Shigeo Kawata (Utsunomiya University) and Prof. Kiyoshi Nishikawa (Kanazawa University) for valuable discussions and useful comments. 7. References [1] NAREGI: National Research Grid Initiative, [2] RENKEI: REsources linkage for E-scIence, [3] S. Matsuoka, S. Shimojo, M. Aoyagi, S. Sekiguchi, H. Usami, K. Miura, Japanese computational grid research project: NAREGI, Proceedings of the IEEE, VOL.93, NO.3, pp , March [4] H.Usami, H.Kanazawa, S.Kawata, A Problem Solving Environment based on Grid Services : NAREGI-PSE,IFIP Working Conference Proceedings, WoCo9 : Grid-based Problem Solving Environment, Springer,pp ,July 2007 [5] I. Foster, H. Kishimoto, et al. The Open Grid Services Architecture, Version 1.5, 2006, [6] I. Foster, C. Kesselman, and S. Tuecke, Anatomy of Grids, International J. Supercomputer s, 15(3), 2001, [7] EU DataGrid, VOMS Architecture version 1.1, 2003, [8] K. Fukui, Contents Service Specification 1.0, 2006, [9] ACS-WG: [10] OGF: Open Grid Forum [11] S. Matsuoka, M. Hatanaka et al., Design and Implementation of NAREGI Super-Scheduler based on the OGSA Architecture, Journal of Computer Science and Technology, 21, 4, (2006),