Researchers in the Netherlands are being granted the opportunity to take part in a grand HPC experiment over the coming year as the limits of BiG Grid are pushed into the cloud. If the full test is a success, this could mean that there will be a significant number of similar efforts in coming years from other national and international grid and research organizations.
As one of the national grid projects in Europe, BiG Grid in the Netherlands hosts “four core centers providing large scale data storage and compute facilities with over twelve distributed seed clusters for Life Sciences, and supporting more than 35 research communities.” Founded in 1970, one of the critical core centers, SARA, is a national high-performance computing and e-Science support center as well as a supernode in the International Science Grid. SARA currently supports researchers with state-of-the-art integrated services, facilities and infrastructure as well as advanced networking, storage, visualization and broader e-Science services.
In 2009, SARA conducted a small-scale pilot experiment with five groups of scientific users to explore the possibilities of cloud computing in an HPC environment. The experiment proved to be a success and BiG Grid decided that it was time to usher in a new phase for the Dutch scientific community and began to offer a still-experimental but much larger-scale offering of HPC cloud as a service — the catch, of course, is that this is only open to members of the Dutch scientific community.
The new HPC cloud environment will provide researchers with a chance to operate within their very own virtual private HPC cluster that can host full individual configuration that will operate according to each scientific team’s needs. The most attractive part of the offer for those who are selected is, of course, on-demand scalability. Participants will be able to start from images or create their own cluster from the bottom-up with the added bonus that users can create a copy of their current software environment (from small or personal machines) and weave it into an HPC cluster operating within the cloud without any expensive rewriting or dramatic changes between their development and production environments.
It won’t just be SARA evaluating the success of the HPC cloud service, of course, the large bevy of researchers from the Netherlands lining up to take part in the experiment will have their eyes peeled for challenges presented by the shift to the cloud including possible performance strains, for example. Following the experimental stage of migration and actually application use in the cloud it will no doubt be fascinating to read what issues emerged for both the infrastructure providers and the scientific community.
While the group states that this experimental phase is open to all Dutch scientific researchers, there are some special members of the community that will be offered prime consideration. As the announcement reveals, the SARA team is particularly interested “in applications which are difficult or nearly impossible to run o the existing HPC platforms (Huygens, Lisa, Grid) but do run in one’s local environment.”
If SARA’s experiment with a larger test of more diverse and numerous users is a success, it seems that this might prove to be a valuable proof of concept for members of other scientific HPC communities around the world. In fact, even if it’s not a total success and there are rampant complaints about any number of hurdles for either side, it’s still a success in that the program is setting the course for other national and large-scale research institutions. The challenges the participants faced can be clearly mapped out and solutions to the barriers can be addressed in a manner that is focus and specific to the needs of scientific HPC users who want to take their capability to the next level.
Recently, HPC in the Cloud spoke with two of the leaders of the SARA cloud experiment, Tom Visser and Floris Sluiter. The following material includes some highlights from the discussion.
Currently, we are involving the dutch scientific community to evaluate this service. At SARA we are convinced that the involvement of the community in the development of new types of infrastructure and services is a key-succes factor.
We are currently only offering this evaluation service to the dutch community, and people affilitated with them. We have received international requests for access in this phase, that we currently cannot accommodate because of the scale of the infrastructure and of funding structures. We have set up partnerships with sister institutes around the globe with whom we exchange experiences and development efforts. This has already lead to the SARA developed graphical management interface (called ONE-MC) for OpenNebula, that is available in open source for the community. We will continue to actively share insights, experiences and results.
What are some of the scientific applications that could be selected for this beta–what types of submissions are you seeing?
We are open for all kinds of suitable applications from the dutch scientific community. We are especially interested in applications which are difficult or near impossible to run on our existing HPC platforms (Supercomputers, Compute clusters, GPU-clusters, or Grids), but do run on their local systems. For example applications that need specific and custom libraries that are difficult to offer and maintain in a shared environment.
We don’t want to limit the kind of applications that users can evaluate, we have already seen new scientific approaches evolving from this new infrastructure. However we will have to make a selection because of the limited scale of the current infrastructure. We strive for a good mix of different usage models, i.e. cloning laptops, large databases, simulation clusters, data-mining, hybrid HPC, virtual private networking, etc.
Using cloud computing in our HPC setting has already elicited new modes of scientific research approaches. For example: a scientist will be able to develop and finetune models on his own workstation. This workstation can be cloned and started as a virtual cluster in the HPC cloud. Now the scientist can run these models seamlessly on an HPC system.
Describe in some detail the this HPC cloud environment and the virtual private cluster–what tools and technologies are you using in particular and what level of configurability do you provide for researchers?
The cloud is hosted on a cluster with 128 cores with the following characteristics:
* 16 compute nodes with dual quad-core CPUs
* backupped storage: 100 TB
* Host Software
o Virtual machine software: KVM
o Multicore/multiprocess is possible, also MPI and OpenMP.
o Virtual Private Compute Cluster(c): starting multiple VMs in their own private network (vlan)
In this beta phase we strive to offer a small production grade environment and we are continuously improving and expanding the technical environment.
The management software we use is OpenNebula and we have developed a web-based userinterface (ONE-MC) on top of that. This software is also available under an open source license. As a community support system we use redmine.
We have developed additional software to manage clusters of virtual machines to enable Virtual Private Compute Clusters.
Security is, particularly in a scientific setting, a major concern. We’ve put various mechanisms in place to assure that users are protected from the outside world, other users and vice-versa.
Users are limited as little as possible, they will have their own Virtual Private Compute Cluster, that can be configured from scratch. They can actually start with an install cd of the operating system of their choice. It is also possible to start by uploading their own pre-configured virtual machine image. As a service we provide configuration templates and provide a community repository for virtual machines. In this repository people can share their own configuration templates and virtual machine images.
Because the HPC infrastructure and computing environment is fully configurable on demand to specific needs, the user can save time and effort porting their applications to a specific HPC platform. In some cases porting can be impossible because the source code is unavailable to the user. Then especially virtualisation can provide a solution. In general we are convinced that a shorter time from scientific question to computational solution will be facilitated by the use of our HPC Cloud.
We are convinced that HPC cloud computing provides a flexible solution to scientist and provides added value to the HPC ecosystem.