eResearch NZ 2013
NeSI is proud to be a gold sponsor for eResearch NZ 2013, a national conference that brings together researchers and technology practitioners to explore the role of technology to enhance research. The team have contributed a wide variety of talks, workshops, and posters to this years event, and also have a booth, making it very easy to discover more about NeSI related platforms, services, and to meet the team.
This page outlines the formal contributions that NeSI staff are making to the conference.
Markus Binsteiner and Sina Masoud-Ansari are explaining some of the ways that NeSI staff look to support researchers who have less confidence with HPC.
In this presentation, we outline our approach to new users, consulting with them to find out what their requirements are and recommending the appropriate tools for what they try to do and, equally important, what their level of HPC-literacy is. Concluding we will talk about some of the infrastructure we have built to be able to support this group of users and give some examples on how we have used it to connect researchers to our compute infrastructure.
Aaron Hicks is delivering a presentation on some of the devops tools which help drive the backend of NeSI's services.
Just as with laboratory technicians, there is a risk that a SysAdmin will produce a Black Box, into which Inputs go and from which outputs come, but this is not science. If the box can not be replicated, nor can the results, and if your results can not be replicated, what science have you really achieved?
Nick Jones, NeSI's Director, is participating in panel with several of New Zealand's research infrastructure providers.
The country now possess several national e-infrastructures. Each has its differences in the way they are constituted, funded, how researchers gain access and the communities they serve. Leaders from these e-infrastructures will come together to discuss their respective roles, reflect on experience and discuss possibilities on the road ahead.
Sina Masoud-Ansari, Gene Soudlenkov and Jordi Blasco will be attempting to compare, from a researcher's prospective, how the NVIDIA CUDA and Intel Phi coprocessor hardware and programming models differ.
Manufacturers including NVIDA, AMD and most recently Intel have made significant investments in developing specialised hardware and software to improve the performance of demanding applications by providing support for large scale parallelism. In this talk, we provide an overview of the architectural differences between the NVIDIA K20X (Kepler) and the Intel Phi 5110P (Knight's Corner) and provide insights into how to achieve good performance on both platforms. In particular, we aim to highlight key differences that proficient programmers should consider when choosing the best architecture for their research problem and for those with less experience, we show how using simplified programming frameworks such as OpenACC can help improve code performance, regardless of the underlying architecture.
Markus Binsteiner and Martin Feller are presenting on how researchers can ship data between collaborators.
This presentation discusses how to efficiently transfer data between sites. Unfortunately transferring data is not always trivial, it very much depends on the size of the files to be transferred, how many of them are there are, where it is located, and where it is supposed to go. Also: who is allowed to see it. We present an overview of several options that are available for researchers in New Zealand. Transfer protocols we will compare include: samba shares, ssh/scp/sftp, irods, and gridftp.
Céline Cattoën-Gilbert and Vladimir Mencl are supporting the work of Stuart Charters discussing transferring 4TB of data between NZ and USA.
We will describe our experiences of transferring 4 TB of data from Oak Ridge National Lab, US to the New Zealand eScience Infrastructure at BlueFern Supercomputing Centre, University of Canterbury.
We will outline the steps we had to go through to gain access to the data and to perform an initial trial using the publically available Data Transfer nodes located at various US sites.
Michael Uddstrom and Mark Cheeseman are running a two day workshop, providing tutorials, talks, and discussions on HPC scientific applications. The workshop will aim to share and discuss insights and experiences in applying high performance computing, focused in particular on the scientific software related considerations in achieving this work.
High performance computing (HPC) has the potential to revolutionise New Zealand research. Many of the lessons for applying HPC to research span research disciplines. As New Zealand’s adoption of HPC matures at differing rates across communities and institutions, an opportunity exists to hold an inter-disciplinary event alongside eResearch NZ 2013 to spread knowledge between practitioners and to strengthen the bonds within the HPC community. The workshop will build upon a successful event in 2012 to become a key meeting place for New Zealand’s HPC community. It has the following objectives:
- provide an opportunity to meet researchers leading their fields in the application of HPC
- provide a forum for meeting and talking with colleagues using scientific applications on HPC
- support a broad community of researchers who depend on HPC applications
Céline Cattoën-Gilbert is providing a tutorial on the use of the Octave programming language within HPC. In particular, there will be an overview of the POWER7 cluster, available to New Zealand researchers through NeSI's access process.
Learn how you can prototype even larger modelling problems in parallel or using distributed memory with Octave in a High Performance Computing environment. Without the cost restriction of MATLAB licenses, it is possible to scale some problems up to hundreds of processors in some circumstances.
Sung Bae will be talking through the use of Cython, mpi4py and other tools to achieve near-native performance from Python, a highly productive programming language.
Despite these advantages, Python is not generally considered to be a good choice for computationally intensive research because it is an interpreted language and slow. It is common to speed up Python code by tens or hundreds times by rewriting the code in C.
However, there are alternatives to rewriting Python code, and we have found that the combination of Cython and mpi4py is a cost-effective solution for Python-based research project codes that need computational speed and scalability.
Jordi Blasco will be introducing the Intel Phi coprocessor, which can greatly increase the speed of computation. The coprocessors will be available for New Zealand researchers through NeSI shortly.
The Intel Xeon Phi coprocessor is suited for massive parallel applications that feature a high ratio of computation to data access. It is composed of up to 61 CPU cores and each core is capable of switching between up to 4 hardware threads on round-robin scheduling, resulting in a total of up to 244 hardware threads available. Each core consists of an slow processor with few extension capabilities (like pentium 4), dual-issue x86 pipeline, a local L1 and L2 cache, and a separate vector processing unit (VPU).
- Pan infrastructural support, Ben Roberts and Gene Soudlenkov, discusses the researcher support that the Centre for eResearch's team within the Pan cluster.
- Integrity and Security of the Pan Cluster, John Spires, Gene Soudlenkov and Yuriy Halytskyy, explains the protections that are in place to ensure the integrity of data
- Evolution of Pan, Rob Borrowes, Marcus Gastafsson and Yuriy Halytskyy.
- New Zealand eResearch Infrastructure, Dan Sun, Michael Uddstrom and Marcus Gastafsson, provides an overview of what NeSI is and what it offers to the research community.