Irish Tier-2 HEI service now open on Fionn – the ICHEC Condominium Service 2013-2016Posted on 6 February 2014
ICHEC is glad to announce the opening of its popular Tier-2 service, known as the ICHEC cluster Condominium shared service on the national supercomputer Fionn.
The transfer formally took place on 1st December 2013, with a total of 303 Condominium users from 7 HEIs transferred from their own institutional Condominium on Stokes to the new Condominiums on Fionn.
This service replaces the very popular service on Stokes, through which over 28 million core-hours have been used by 321 researchers from 6 Irish research institutions.
The tight integration of Tier-2 access (institutional, typically by Research Office request) and Tier-1 (national service, through peer review) has demonstrated significant benefits to researchers, their institutions and funding agencies, both in terms of the quality of service and the unmatched value for money through the effective leveraging of prior Exchequer investments.
What is a Tier-2 Service?
The Branscomb Pyramid (see below) provides a conceptual framework for understanding and thinking about the provision of an optimal mix of HPC resources. Although the level of resources at each level of the pyramid varies from time to time, the framework provides a convenient and elegant way of thinking about the scale, costs, and uses of HPC infrastructure. Branscomb argued that an ecosystem of compute resources is needed, ranging from high performance workstations, through small clusters, supercomputers found in national centres, and high-end facilities in large international centres. As these performance levels quoted in Figure 1 double every 18 months (in accordance with Moore’s law), the current performance spread ranges from a few to tens of Petaflops (Pflops1) for so-called Tier-0 sites such as those operated by the four PRACE hosting partners, to 100s of Tflops to a few Pflops in National HPC Centres, 10s Tflops for University Centres, to Tflops for powerful workstations.
Figure 1: The Branscomb Pyramid for Computing for 2006 [see also NSF Blue Ribbon Panel on High Performance Computing, August 1993]. (Note that in 2014, leadership class supercomputers are 10s Pflops; large-scale supercomputers 100s Tflops to 1s Pflops; and medium-scale resources 10s of Teraflops)
In Ireland, access to Tier-0 resources has been facilitated through ICHEC by applying to internationally EU competitive programmes, such as PRACE and the US DoE INCITE programme. The current National HPC supercomputer, Fionn, has been rated at 147 Tflops and provides Tier-1 access. Tier-2 is provided through a combination of local resources at HEIs as well as through ICHEC’s Tier-2 cluster Condominium shared service, which provides a cost effective alternative to local cluster ownership, while offering superior functionality.
What is the Cluster Condominium shared service?
ICHEC is fully aware of the need for ‘local’, institutionally-accessed resources, and in response to this need has operated a Tier-2 service for HEIs, known as the ICHEC cluster Condominium shared service (or Condo in short) since its establishment in 2005. ICHEC has actively promoted the development of a centralised shared services model, in part to facilitate the most effective means of sharing scarce human resources (expert systems administrators), and in part also to ensure optimum use and value for money of expensive hardware. Over time, the HEIs have increasingly exploited the benefits of this model.
ICHEC now runs ‘local’ HPC systems as part of its centralised compute services for seven of the higher education/research institutions. These ‘local’ resources (in fact additional ‘partitions’ added to the main ICHEC computer) are known as condominiums and are an outstanding example of the benefits of user-friendly, customer-focused shared services. These condos represent cost-effective alternatives to individual institutions procuring, buying and managing their own HPC resources for small systems (say of the order of 100+ cores). This development will continue to form an important part our future strategy for the provision of HPC resources nationally, given the de facto formalisation of the condominium shared services model as a primary vehicle to provide ‘local’ (Tier-2) services.
Up to 30% of these central resources will be available to the research institutions as ‘local’ condominium clusters, with access controlled by a local administration (usually the IT Services and Research Office working together to approve local access). In signing up to a centralised model for resources, the HEIs achieve significant savings, and researchers’ access to a high availability, highly tuned infrastructure providing a trivial migration path to the National HPC Service.
We are very confident that this shared service condominium model will continue to develop and the model remains open to adoption by remaining HEIs.
What are the benefits of the ICHEC Condo Tier-2 Service?
In terms of costs, it is clear that the ICHEC Condo Tier-2 Service brings significant savings in terms of national funding for both the funding agencies and for the HEIs. The savings achieved via the shared services approach are easily identified:
The research institutions will save on procurement overheads for the equipment.
ICHEC’s engagement with the vendor provides the HEIs with a cluster at a significantly lower cost compared to buying similar equipment directly as smaller stand-alone systems.
ICHEC guarantees that the institution obtains its own institutional resources, managed by highly experienced ICHEC staff; local human resources are not diverted to managing the cluster.
ICHEC central resources are administered to near mission-critical levels (currently >99% availability).
Significant savings are also made on storage and cluster infrastructure (management nodes, login nodes, interconnect, software licenses, etc.) by leveraging the economy of scale.
Hosting and systems administration overheads are costed centrally leading to significant savings. Apart from the initial capital cost of the condo and storage, the only additional costs for the institution are the monthly electricity costs charged pro rata to the size of the condo.
Central purchasing and installation of up-to-date, tuned tools, libraries, application software.
Natural progression from Tier 2 (“local”) to Tier 1 (“national”), facilitating access to internationally competitive resources (e.g., PRACE Tier-0 and 1, subject to peer review).
Condos persist for the duration of the underlying ICHEC national research HPC cluster.
Greater utilisation of their resources, where institutions use the allocation model rather than the partition model for access.
In summary, therefore, ICHEC believes that the condominium shared services model is a very attractive one for institutions and funding agencies alike. It leads to significant saving of costs, management and administrative overheads. It represents in ICHEC’s view, a realistic and effective road map for the provision of high-end compute resources for the national research community and one that meets the objectives and needs of that community. Condos are not a substitute to the National Service – but are a more cost-effective and flexible alternative to direct ownership of small clusters. It is also worth noting that this model was also strongly endorsed by panels of international experts convened by SFI as part of its mid-term review of ICHEC (Q4-2011) and by the HEA for its “Review of the Provision of High Performance Computing in Ireland” in 2012.
Which Irish Higher Education Institutions are participating in the programme?
Condo participants on Fionn include:
Dublin City University (DCU)
Dublin Institute for Advanced Studies (DIAS)
National University of Ireland, Galway (NUIG)
National University of Ireland, Maynooth (NUIM)
Tyndall National Institute (TNI)
University College Dublin (UCD)
University of Limerick (UL)
The programme is open to all HEIs in the Republic of Ireland and Northern Ireland.
How can I access this service?
Participating HEIs have appointed a designated member of their institution as Condo Point of Contact (see list at http://www.ichec.ie/services/alternate_access). Please contact this person to request access. Assuming he/she authorises your request, access is normally granted within the next day or two.
1 A flops (for FLoating-point Operations Per Second) is a measure of computer performance.
|PRACE Autumn School 2014
25-28 Nov, Athens, Greece
|DL_POLY tutorial in Cork
DL_POLY tutorial in Cork
|Volcanic Ash Cloud Alert
NUI Galway's C-CAPS and ICHEC react rapidly in response to latest alert
Molecular-Scale Transition Metal Oxide Nanocluster Surface-Modified Titanium Dioxide as Solar-Activated Environmental Catalysts
|ICHEC and NewsWhip take on Viral stories
A collaboration conducted to predict how viral a story will be