Abstract
Within a few short years, we can expect to be dealing with multi-million-thread programs running on million-core systems [16]. This will no doubt stress the contemporary HPC software model which was developed in a time when 512 cores was a large number. Historical approaches have been further challenged by the increased desire of developers and end users for supercomputer light weight kernels (LWKs) to support the same environment, libraries, and tools as their desktops. As a result, the emerging workloads of today are far more sophisticated than those of the last two decades when much of the HPC infrastructure was developed, and feature the use of scripting environments such as Python, dynamic libraries, and complex multi-scale physics frameworks. Complicating this picture is the overwhelming management, monitoring and reliability problem created by the huge number of nodes in a system of that magnitude.
We believe that a re-evaluation and exploration of distributed system principals is called for in order to address the challenges of ultrascale. To that end we will be evaluating and extending the Plan 9 [21] distributed system on the largest machines available to us, namely the BG/L [28] and BG/P [10] supercomputers. We have chosen Plan 9 based on our previous experiences with it in combination with previous research [17] which determined Plan 9 was a "right weight kernel", balancing trade offs between LWKs and more general purpose operating systems such as Linux. To deal with issues of scale, we plan on leveraging the use of the high-performance interconnects by system services as well as exploring aggregation as more of a first-class system construct -- providing dynamic hierarchical organization and management of all resources. Our plan is to evaluate the viability of these concepts at scale as well as create an alternative development and execution environment which compliments the features and capabilities of the existing system software and run time options. Our intent is to broaden the application base as well as make the system as a whole more approachable to a larger class of developers and end-users.
- G. Ammons, J. Appavoo, M. Butrico, D. D. Silva, D. Grove, K. K. an Orran Krieger, B. Rosenburg, E. V. Hensbergen, and R. W. Wisniewski. Libra: A library operating system for a JVM in a virtualized execution environment. To appear in the Proceedings of VEE 2007, 2007. Google Scholar
Digital Library
- M. Burrows. The Chubby lock service for loosely-coupled distributed systems. In Proceedings of the 7th Symposium on Operating System Design and Implementation (OSDI '06), pages 335--350, Berkeley, CA, USA, 2006. USENIX Association. Google Scholar
Digital Library
- S. Chakravoty and et. al. Hpc-colony: services and interfaces for very large systems. ACM SIGOPS Operating Systems Review, 40(2), April 2006. Google Scholar
Digital Library
- Coraid. The linux storage people. http://www.coraid.com.Google Scholar
- S. Corporation. Throughput Computing: Changing the economics and ecology of the data center with innovative SPARC technology. Technical report, Sun Corporation, November 2005.Google Scholar
- R. C. Emil Sit, Josh Cates. A dht-based backup system. In In Proceedings of the First IRIS Student Workshop, 2003.Google Scholar
- C. Engelmann and et. al. Molar: Adaptive runtime support for high-end computing operating and runtime systems. ACM SIGOPS Operating Systems Review, 40(2), April 2006. Google Scholar
Digital Library
- E. V. Hensbergen. P.R.O.S.E.: partitioned reliable operating system environment. SIGOPS Oper. Syst. Rev., 40(2):12--15, 2006. Google Scholar
Digital Library
- E. V. Hensbergen, J. McKie, C. Forsyth, and R. Minnich. Night of the lepus: A plan 9 perspective on blue gene's interconnects. In In Proceedings of the second annual international workshop on Plan 9, 2007.Google Scholar
- IBM BlueGene Team. Overview of the Blue Gene/P project. IBM Journal for Research and Development, 52(1/2), January 2008. Google Scholar
Digital Library
- J. A. Kahle, M. N. Day, H. P. Hofstee, C. R. Johns, T. R. Maeurer, and D. Shippy. Introduction to the Cell multiprocessor. IBM Journal of Research and Development, 49(4/5):589--604, 2005. Google Scholar
Digital Library
- S. Kelly and R. Brightwell. Software architecture of the light weight kernel, catamount. In Proceedings of the 2005 Cray User Group Conference, 2005.Google Scholar
- L. Lamport. The part-time parliament. ACM Trans. Comput. Syst., 16(2):133--169, 1998. Google Scholar
Digital Library
- L. Lamport and M. Massa. Cheap Paxos. In DSN '04: Proceedings of the 2004 International Conference on Dependable Systems and Networks (DSN'04), page 307, Washington, DC, USA, 2004. IEEE Computer Society. Google Scholar
Digital Library
- B. W. Lampson. How to build a highly available system using consensus. In WDAG '96: Proceedings of the 10th International Workshop on Distributed Algorithms, pages 1--17, London, UK, 1996. Springer-Verlag. Google Scholar
Digital Library
- M. Leininger. Application and algorithm challenges for future petascale to exascale computing architectures. In OASCR AMR Future Architectures Panel, 2007.Google Scholar
- R. Minnich and et. al. Right-weight kernels: an off-the shelf alternative to custom light-weight kernels. ACM SIGOPS Operating Systems Review, 40(2), April 2006. Google Scholar
Digital Library
- R. G. Minnich, A. Mirtchovski, and L. Ionkov. XCPU: a new, 9p-based, process management system for clusters and grids. In Cluster 2006, 2006.Google Scholar
- R. Pike and et. al. Plan 9 - the documents (volume 2). http://plan9.bell-labs.com/sys/doc.Google Scholar
- R. Pike and et. al. Plan 9 - the manual (volume 1). http://plan9.bell-labs.com/sys/man.Google Scholar
- R. Pike, D. Presotto, S. Dorward, B. Flandrena, K. Thompson, H. Trickey, and P. Winterbottom. Plan 9 from Bell Labs. Computing Systems, 8(3):221--254, 1995.Google Scholar
- R. Pike, D. Presotto, S. Dorward, D. M. Ritchie, H. Trickey, and P. Winterbottom. The Inferno operating system. Bell Labs Technical Journal, 2(1), Winter 1997.Google Scholar
- S. Quinlan and S. Dorward. Venti: a new approach to archival storage. In In Proceedings of the Conference on File and Storage Technologies, 2002. Google Scholar
Digital Library
- T. Sanders. Meet Larrabee, Intel's answer to a GPU. Vunet.com, 2007. http://www.theinquirer.net/default.aspx?article=37548.Google Scholar
- M. Sottile and R. Minnich. Analysis of microbenchmarks for performance tuning of clusters. In Proceedings of Cluster 2004, 2004. Google Scholar
Digital Library
- M. J. Sottile and R. G. Minnich. Supermon: A high-speed cluster monitoring system. In CLUSTER '02: Proceedings of the IEEE International Conference on Cluster Computing, page 39, Washington, DC, USA, 2002. IEEE Computer Society. Google Scholar
Digital Library
- I. Stoica, R. Morris, D. Karger, M. F. Kaashoek, and H. Balakrishnan. Chord: A scalable peer-to-peer lookup service for Internet applications. ACM SIGCOMM 2001, pages 149--160, August 2001. Google Scholar
Digital Library
- The BlueGene/L Team. An overview of the BlueGene/L supercomputer. In ACM Supercomputing Conference, 2002. Google Scholar
Digital Library
- K. Thompson and G. Collyer. The 64-bit standalone plan 9 file server. http://plan9.bell-labs.com/sys/doc/fs.pdf.Google Scholar
- Uriel. 9p implementations. http://9p.cat-v.org/implementations.Google Scholar
- D. Wallace. Compute node linux: New frontiers in compute node operating systems. In In Proceedings of the Cray User's Group, 2007.Google Scholar
Index Terms
Holistic aggregate resource environment
Recommendations
Spatial Data Resource Index in Grid Environment
GCC '09: Proceedings of the 2009 Eighth International Conference on Grid and Cooperative ComputingGrid is a promise technology which could easily share distributed heterogeneous computing resources. It brings together geographically and organizationally dispersed computational resources. This paper presented a framework that aggregated the spatial ...
The Grid Resource Broker workflow engine
2nd International Workshop on Workflow Management and Applications in Grid Environments (WaGe2007)Increasingly, complex scientific applications are structured in terms of workflows. These applications are usually computationally and-or data intensive and thus are well suited for execution in grid environments. Distributed, geographically spread ...
Hyper-heuristic Based Resource Scheduling in Grid Environment
SMC '13: Proceedings of the 2013 IEEE International Conference on Systems, Man, and CyberneticsAn efficient management of the resources in Grid computing crucially depends on the efficient mapping of the jobs to resources according to the user's requirements. Grid resources scheduling has become a challenge in the computational Grid. The mapping ...






Comments