Distributed Thinking

Ian Foster notes a project at University of Notre Dame near Chicago which is distributing research computing facilities in order to provide heat to campus buildings. This is an interesting trade-off: on the one hand, the distribution of resources means that the waste heat is put to good use; on the other hand, it's possible that each distributed installation is less efficient that a good-quality centralised machine hall. This is another example of where we need good quality models and well-measured example deployments to help us decide which approaches give the best results.

This week, I attended Healthcare Computing 2008 to get an update on the current state of e-health in the UK and to explore how grid technology can contribute. Health Informatics is a broad subject and it isn't possible to engage with the whole field, but I see three main areas of potential engagement. The path is most followed by the academic community to date is that of linking together data used in clinical trials or in health research. This is a natural fit for the e-science community as it extends existing work on secure access to distributed research data. The medical world imposes more security constraints, which adds academic interest, but is otherwise familiar to the e-scientists. It is also of a scale that is manageable in research projects. Successful projects include Psygrid , which is now deployed across the NHS research centres in mental health and in bioinformatics. So this is the first area of engagement. A natural question is whether this experience with rese...

There has been some rather confused coverage in the press about the grid infrastructure that supports the Large Hadron Collider at CERN. The Times has an article that cast the grid as a "superfast internet", with the emphasis on the high-bandwidth links that have been laid to support the LHC data dispersal. It also talks about the numbers of servers connected to the LHC grid, but without clarifying the distinction between bandwidth and processing power. It also implies that the LHC grid is the only grid, which perhaps we can forgive the journalists for, as plenty of technical people still refer to "the grid" as if there were only one. A Yahoo article , taken from Sky News , goes rather further, claiming that "the internet, as we know it, could be obsolete within a decade". The phrase, "as we know it", lends a wonderful vagueness to the claim. The article goes on to say that the Grid was the brainchild of CERN, which of course is an exaggera...

Ian Foster pointed readers of his blog at a good article about CaBIG, the Cancer Biomedical Informatics Grid. This major project is led by the US National Cancer Institute and also involves Cancer Research Labs in the UK, with the aim to share data between cancer researchers. The UK's OGSA-DAI system is a major component of the deployed system. It's particularly interesting to note that the system includes support for clinical trials. Clinical trials are time-consuming and expensive, so many people want better systems for managing them and a number of grid projects are tackling this area. PsyGrid is one such project in the UK. They're not exactly blowing their own trumpet about this, but their system is being used as the studies and trials platform for the Mental Health Research Network, and has been selected by the National Institute of Health Research (NIHR) as the platform for providing electronic data collection services to support studies and trial across the ...

I am very pleased with the line-up for our webinar on May 7th. We will continue our investigation into best practice for running energy efficient data centres, following our successful webinar last October. This time we will have an emphasis on the impact of virtualisation and on how one can model and measure the effectiveness of proposed improvements. Nic Barnes will explain what Merrill Lynch observed when they applied virtualisation on the desktop and in the server room. They found that the benefits were in some cases partially offset by losses elswhere. Nic will demonstrate the importance of measuring real gains in practice. Liam Newcombe will describe the work of the BCS in developing a model for predicting data centre efficiency. This approach will allow managers to plan and evaluate designs in advance of their implementation. Liam will show that the choice of metrics requires careful analysis. As always, viewers will be encouraged to ask questions. Any that we can...

Virtualisation has obvious benefits for much commercial IT, where existing servers often have utilisation rates of 10% or less. It's less clear whether virtualisation is so useful for High-Performance Computing (HPC), where systems are often kept running at utilisation rates above 90%. Nevertheless, there are potential benefits from adopting some aspects of virtualisation in these environments. The question is, do the benefits outweigh the cost in performance? This was the topic of yesterday's workshop on System Level Virtualisation for HPC (which was part of EuroSys 2008 ). The workshop was rather small but I did learn quite a bit. A couple of talks investigated the performance hits of using virtualised OS's instead of addressing the hardware directly. This varied, depending on the type of application; if IO was minimal, the slowdown was minimal too. An "average" slowdown seemed to be on the order of 8%. Samuel Thibault of Xensource looked at ways of usin...