D2 3 Computing e-Infrastructure cost calculations and business _models_vam1-final

e-­‐FISCAL: www.efiscal.eu
EC Contract Number: 283449
14 out of the 28 respondents provide both computing and coordination activities while 13 provide only computing
services. Only one respondent was solely a coordination organisation.
Out of the 28 respondents, 14 (50%) indicated that their institute is only part of NGI/EGI, 3 reported participating
into National HPC infrastructure/PRACE, while another 8 participate in both.
The respondents belong to institutes that have multiple roles ending up in a vast number of combinations. The
majority of institutes participating in e-­‐FISCAL survey (20 out of 28) are NGI Resource Centres (CPU, storage, etc.),
while another 15 out of 28 are HPC centres. As a result, there is no easy separation between HTC and HPC centres,
especially when comparing the high-­‐end HTC ones with the low-­‐end HPC ones 47 . Details about the characteristics
of the sample are found in Appendix 7.4.
4.2 Research question 1: Analyse the cost structure of HPC and HTC centres
In order to analyse the cost structure of HPC and HTC centres, we used the cost model presented in section 3.2.1.
The analytical steps in cost calculations are presented in a comprehensive and transparent way in Appendix 7.3.
The descriptive statistics of our sample are presented in detail in Appendix 7.4. The project aimed at calculating
the costs for both 2010 and 2011. In this deliverable, we focus on the 2011 cost data as the cost data of 2010
seemed less relevant at the time the report was written (January 2013) 48 . However, it should be noted that costs
between 2010 and 2011 are decreasing in-­‐line with the trend of lower hardware prices and better overall
efficiency. In order to get a multifaceted view of the cost structure of HPC and HTC centres, we treat input data
under three different option cases.
Option 1: Basic Case: We run the e-­‐FISCAL costing model by using as inputs the average and the median 49 values
of all necessary input data from the total e-­‐FISCAL sample (i.e. the average and the median value of acquisition
cost for CPUs for the total sample, the average and median depreciation rates of the total sample etc.). Even
though this procedure may not adequately account for the existence of economies of scale and does not
correspond to the specific circumstances of each and every participant in the study, it makes a fair approximation
of the “average” case. Additionally, the core unit of analysis permits the presentation of results in a way that is
easy to understand and follow. It is a therefore a reference point that could be used for high-­‐level comparisons.
The results are presented as follows (detailed calculations are found in Appendix 7.3):
47 This was a comment raised by the reviewers of one of the intended publications of the project (for the eChallenges 2012
conference) and was also raised with the e-­‐Infrastructure community during the e-­‐IRG workshop. As there are several
definitions of these two terms, some of which are not satisfactory (as the Wikipedia one), the project decided to raise the
attention of the e-­‐Infrastructure community and especially EGI and PRACE. In the Terms and Definitions section we have
adopted the EGI definition.
48 A thorough analysis of the 2010 cost information and break down is found in D2.2: Computing e-­‐Infrastructure cost
calculation at national and European level found in http://www.efiscal.eu/deliverables.
49 The median is described as the numeric value separating the higher half of a sample, a population, or a probability
distribution, from the lower half.
e-­‐FISCAL : Financial Study for Sustainable Computing e-­‐Infrastructures
Deliverable D2.3 – Computing e-­‐Infrastructures cost estimation and analysis – Pricing and
Business models
43