Green Desktop-Grids: Scientific Impact, Carbon Footprint, Power Usage Efficiency

Desktop Grids take their place in the e-Science distributed computing infrastructure scaling into the millions of PCs. Desktop-Grids collect CPU cycles from PCs contributed by donors, by volunteers who are willing to support science and research. The Green key advantage of Desktop-Grids over service Grids and data centers based on clusters of servers is the minimal heat density. Compute Clusters without energy intensive aircondition would run into thermal disaster within minutes. PCs participating in Desktop-Grids usually do not make use of any aircondition. In return and with raising energy prices, data centers have implemented lower costs air-conditioning means like e.g. free cooling, improving their thermodynamic-efficiency and PUE rating. This paper, based on [24], investigates whether Desktop-Grids still have a Green advantage over Service-Grids and describe several distinct Green Methodologies to optimize compute unit specific energy consumption. Green-IT metrics as Carbon Footprint and PUE are analyzed for their relevance and applied to Desktop-Grids. Pragmatic implementation steps to Green-Desktop-Grids are described. 1. Desktop-Grid: Scientific Impact of large scale DCIs. During the past years, volunteer Desktop Grids have become a regular part of the computational infrastructure for e-Science. Although they already form an impressive computational power, the EDGeS infrastructure for instance connects about 150.000 computers in Desktop Grids to the European Grid Infrastructure (EGI), this is only the beginning, as there are hundreds of millions of computers, alone in Europe that could be connected. Desktop Grids take their place in the e-Science distributed computing infrastructure scaling into the millions of PCs [23]. Compute time harvested from resource owners, typically individuals at home but also institutions and companies, does not request large upfront investment by the scientist; it is a low cost approach towards significant scientific output. Typically implemented using BOINC [17], sometimes XtremWeb [19], OUR-Grid [49], or other packages, Desktop-Grids are found among the largest Distributed Compute Infrastructures (DCI) [1]. Also known as Volunteer-Computing, Desktop-Grids have been around since the very early days of Grid computing [18]. An outstanding illustration of Desktop-Grid scientific impact was delivered these days by Einstein@Home. The new binary radio pulsar J1952+2630 was detected in data recorded at the Arecibo telescope back in 2005 [35]. The aggregations of so many machines result in significant performance well beyond Petaflop/s for selected applications. For example: BOINC network averages about 5.1 Petaflop/s as of April 21, 2010 [2]. Key difference to service Grids like EGEE (now EGI) is the voluntary character of the resources citizens contribute their PCs compute time to the Desktop-Grid projects in order to support scientific challenges of their choice. The FP7 project DEGISCO [20] supports Desktop-Grid deployments in and beyond Europe, especially countries that strongly collaborate with the European Union. DEGISCO recently published a first version of the Desktop Grids for eScience Road Map [21, 22], a guide to prepare and implement successful Desktop-Grid roll-outs, second release scheduled for June 2011. DEGISCO is accompanied by the EDGI project that continues to maintain and further develop the 3G-Bridge [3], a gateway transparently connecting gLite, Unicore, and KnowArc based infrastructures (Service-Grids) to Desktop-Grids by automated translation of the job-languages. Sustained development and success of this combination of Desktop-Grids and Service-Grids has been proven at the EGI User Forum 2011 in Vilnius, Lithuania, by command-line submission of 10,000 jobs to a Desktop Grid from and through gLite, one of the main EGI Grid stacks [26]. Fascinating progress in material science using Desktop-Grids was reported by O. Gatsenko et al at the CGW2010 [27, 28, 29, 30], proving even complex problems to be solvable on a distributed Desktop-Grid platform. 2. The need for Green Desktop-Grids. One core topic of DEGISCO is the energy efficient handling of Desktop-Grid workload and management of resources, provided as configuration advice to Desktop-Grid operators. The need for Green-Desktop-Grids derives from sheer size: Desktop-Grids can aggregate hundredthousands of machines per project. Power consumption of such large amounts of devices should be considered when making use of them. Indeed, when used for computation, energy consumption of PCs (like of any other computer) goes up [5,31]. The contributor, the volunteer, who allows and enables the use of her or his machine, AlmereGrid and VCOdyne SAS, Le Chesnay, France, E-mail: [email protected] AlmereGrid, Almere, Netherlands, E-mail: [email protected]