HotCarbon’2022
Keynote
Ian Schneider (Google, Inc.)
Carbon emissions are an increasingly salient metric for data centers and computer systems, as well as for the software that runs on them. As organizations seek to reduce computer system carbon emissions, we have much to learn about the level of carbon data precision that is necessary to drive effective sustainability actions. Data center carbon emissions come primarily from the electricity used to power computer systems, the embodied emissions of IT equipment, and the emissions associated with data center construction. To reduce carbon emissions, we need data on the key drivers of carbon emissions. However, the available data quality is low, especially for embodied emissions. Furthermore, carbon emissions are generally correlated with overall costs. In many cases, efforts to reduce overall costs already help to reduce carbon emissions, and efforts to directly reduce carbon emissions may be hindered by imprecise carbon data or incomplete coverage. However, there are important exceptions to the general correlation between costs and carbon emissions. For example, specialized hardware like an AI accelerator could increase costs but reduce energy consumption for certain types of software. Moreover, lower-carbon suppliers may charge higher costs for commodity products. How should organizations chart a path towards more precise carbon emission data, and what is the appropriate level of investment in higher-quality data? What are the areas and decisions where operating costs serve as a useful proxy for carbon emissions, and what are the areas where these two criteria diverge?
Session 1: Hardware
Swamit Tannu (University of Wisconsin-Madison); Prashant J. Nair (University of British Columbia)
Amanda Tomlinson, George Porter (University of California, San Diego)
Wenpeng Wang, Victor Ariel Leal Sobral, Md Fazlay Rabbi Masum Billah, Nurani Saoda, Nabeel Nasir, Bradford Campbell (University of Virginia)
Session 2: Networking
Noa Zilberman (University of Oxford); Eve Schooler, Uri Cummings (Intel); Rajit Manohar (Yale University); Dawn Nafus (Intel); Robert Soulé (Yale University); Rick Taylor (Ori Industries)
Agrim Gupta, Ish Kumar Jain, Dinesh Bharadia (University of California San Diego)
Session 3: Metrics
Noman Bashir, David Irwin, Prashant Shenoy, Abel Souza (University of Massachusetts Amherst)
Andrew A. Chien (University of Chicago and Argonne National Laboratory); Chaojie Zhang, Liuzixuan Lin, Varsha Rao (University of Chicago)
Anshul Gandhi (Stony Brook University); Kanad Ghose, Kartik Gopalan (Binghamton University); Syed Rafiul Hussain (Pennsylvania State University); Dongyoon Lee (Stony Brook University); Yu David Liu (Binghamton University); Zhenhua Liu (Stony Brook University); Patrick McDaniel (Pennsylvania State University); Shuai Mu, Erez Zadok (Stony Brook University)
Vaastav Anand, Zhiqiang Xie, Matheus Stolet, Roberta De Viti, Thomas Davidson, Reyhaneh Karimipour, Safya Alzayat, Jonathan Mace (Max Planck Institute for Software Systems (_MPI-SWS)_)
Session 4: Datacenters
Bilge Acun (Meta AI / FAIR); Benjamin C. Lee (University of Pennsylvania, Meta); Fiodar Kazhamiaka (Stanford University); Aditya Sundarrajan, Manoj Chakkaravarthy (Meta); Kiwan Maeng (Meta AI); David Brooks (Harvard); Carole-Jean Wu (Meta AI / FAIR / ASU)
Prateek Sharma (Indiana University)
Thomas Anderson (University of Washington); Adam Belay (MIT); Mosharaf Chowdhury (University of Michigan); Asaf Cidon (Columbia); Irene Zhang (Microsoft Research)
Colleen Josephson (VMware and UC Santa Cruz); Zhelong Pan, Nicola Peill-Moelter, Victor Firoiu, Ben Pfaff (VMware)