Search for: All records

Generative AI, exemplified in ChatGPT, Dall-E 2, and Stable Diffusion, are exciting new applications consuming growing quantities of computing. We study the compute, energy, and carbon impacts of generative AI inference. Using ChatGPT as an exemplar, we create a workload model and compare request direction approaches (Local, Balance, CarbonMin), assessing their power use and carbon impacts. Our workload model shows that for ChatGPT-like services, in- ference dominates emissions, in one year producing 25x the carbon-emissions of training GPT-3. The workload model characterizes user experience, and experiments show that carbon emissions-aware algorithms (CarbonMin) can both maintain user experience and reduce carbon emissions dramatically (35%). We also consider a future scenario (2035 workload and power grids), and show that CarbonMin can reduce emissions by 56%. In both cases, the key is intelligent direction of requests to locations with low-carbon power. Combined with hardware technology advances, CarbonMin can keep emissions increase to only 20% compared to 2022 levels for 55x greater workload. Finally we consider datacenter headroom to increase effectiveness of shifting. With headroom, CarbonMin reduces 2035 emissions by 71%. 

Cloud providers are adapting datacenter (DC) capacity to reduce carbon emissions. With hyperscale datacenters exceeding 100 MW individually, and in some grids exceeding 15% of power load, DC adaptation is large enough to harm power grid dynamics, increasing carbon emissions, power prices, or reduce grid reliability. To avoid harm, we explore coordination of DC capacity change varying scope in space and time. In space, coordination scope spans a single datacenter, a group of datacenters, and datacenters with the grid. In time, scope ranges from online to day-ahead. We also consider what DC and grid information is used (e.g. real-time and day-ahead average carbon, power price, and compute backlog). For example, in our proposed PlanShare scheme, each datacenter uses day-ahead information to create a capacity plan and shares it, allowing global grid optimization (over all loads, over entire day). We evaluate DC carbon emissions reduction. Results show that local coordination scope fails to reduce carbon emissions significantly (3.2%–5.4% reduction). Expanding coordination scope to a set of datacenters improves slightly (4.9%–7.3%). PlanShare, with grid-wide coordination and full-day capacity planning, performs the best. PlanShare reduces DC emissions by 11.6%–12.6%, 1.56x–1.26x better than the best local, online approach’s results. PlanShare also achieves lower cost. We expect these advantages to increase as renewable generation in power grids increases. Further, a known full-day DC capacity plan provides a stable target for DC resource management. 

Traditional datacenter design and optimization for TCO and PUE is based on static views of power grids as well as computational loads. Power grids exhibit increasingly variable price and carbon-emissions, becoming more so as government initiatives drive further decarbonization. The resulting opportunities require dynamic, temporal metrics (eg. not simple averages), flexible systems and intelligent adaptive control. Two research areas represent new opportunities to reduce both carbon and cost in this world of variable power, carbon, and price. First, the design and optimization of flexible datacenters. Second, cloud resource, power, and application management for variable-capacity datacenters. For each, we describe the challenges and potential benefits. 

The rapid growth of distributed energy resources (DERs) is one of the most significant changes to electricity systems around the world. Examples of DERs include solar panels, small natural gas-fueled generators, combined heat and power plants, etc. Due to the small supply capacities of these DERs, it is impractical for them to participate directly in the wholesale electricity market. We study in this paper an efficient aggregation model where a profit-maximizing aggregator procures electricity from DERs, and sells them in the wholesale market. The interaction between the aggregator and the DER owners is modeled as a Stackelberg game: the aggregator adopts two-part pricing by announcing a participation fee and a per-unit price of procurement for each DER owner, and the DER owner responds by choosing her payoff-maximizing energy supplies. We show that our proposed model preserves full market efficiency, i.e., the social welfare achieved by the aggregation model is the same as that when DERs participate directly in the wholesale market. 

Connections across commodity markets create the potential for risk to propagate and for failures to cascade as successive market agents fail. The structure of these networks is, however, often hidden and not directly observable. This article describes methods to uncover this hidden structure and the implications that these hidden connections may have for predicting risk propagation and cascading failures. The results are described in the context of electricity, gasoline, and financial markets. They illustrate the potential of this methodology to help address energy and commodity policy issues and their environmental implications. 

The rapid growth of datacenter (DC) loads can be leveraged to help meet renewable portfolio standard (RPS, renewable fraction)targets in power grids. The ability to manipulate DC loads over time(shifting) provides a mechanism to deal with temporal mismatch between non-dispatchable renewable generation (e.g. wind and solar) and overall grid loads, and this flexibility ultimately facilitates the absorption of renewables and grid decarbonization. To this end, we study DC-grid coupling models, exploring their impact on grid dispatch, renewable absorption, power prices, and carbon emissions.With a detailed model of grid dispatch, generation, topology, and loads, we consider three coupling approaches: fixed, datacenter-local optimization (online dynamic programming), and grid-wide optimization (optimal power flow). Results show that understanding the effects of dynamic DC load management requires studies that model the dynamics of both load and power grid. Dynamic DC-grid coupling can produce large improvements: (1) reduce grid dispatch cost (-3%), (2) increase grid renewable fraction (+1.58%), and (3) reduce DC power cost (-16.9%).It also has negative effects: (1) increase cost for both DCs and non-DC customers, (2) differentially increase prices for non-DC customers, and (3) create large power-level changes that may harm DC productivity. 

Datacenter scheduling research often assumes resources as a constant quantity, but increasingly external factors shape capacity dynamically, and beyond the control of an operator. Based on emerging examples, we define a new, open research challenge: the variable capacity resource scheduling problem. The objective here is effective resource utilization despite sudden, perhaps large, changes in the available resources. We define the problem, key dimensions of resource capacity variation, and give specific examples that arise from the natural world (carbon- content, power price, datacenter cooling, and more). Key dimensions of the resource capacity variation include dynamic range, frequency, and structure. With these dimensions, an empirical trace can be character- ized, abstracting it from the many possible important real-world generators of variation. Resource capacity variation can arise from many causes including weather, market prices, renewable energy, carbon emission targets, and internal dynamic power management constraints. We give examples of three dif- ferent sources of variable capacity. Finally, we show variable resource capacity presents new scheduling challenges. We show how variation can cause significant performance degra- dation in existing schedulers, with up to 60% goodput reduction. Further, initial results also show intelligent scheduling techniques can be helpful. These insights show the promise and opportunity for future scheduling studies on resource volatility. 

Generation type of power plant (e.g. steam, wind) is an important attribute in power grid and energy market studies such as bidding strategy, audit of generation mix, and accounting for load- generation matching. Recently, regional transmission organizations (RTOs) and independent system operators (ISOs) are increasingly redacting a wide range of grid and market data attributes to protect their participants’ business interests. Lack of this information can prevent important power grid research. We propose techniques to infer power plant generation types based on publicly-available market data. We develop and evaluate these techniques on data available from the Midcontinent Independent System Operator (MISO). Evaluation shows successful classification of power plants, achieving 100% precision and 99.5% recall for wind plants, and 91.7% overall accuracy. On the basis of generated power, our classification shows 100% precision and 99.8% recall for wind plants and 93.2% overall accuracy. Our ultimate goal is to generalize to a wide range of RTOs/ISOs. We explore three feature types (bid pattern, capability, and opera- tion), and evaluate their classification value for MISO. We also assess applicability to other RTOs/ISOs based on available market data. These studies inform the efficacy of the features for generation-type inference in other RTOs/ISOs. 

Today's serverless provides "function-as-a-service" with dynamic scaling and fine-grained resource charging, enabling new cloud applications. Serverless functions are invoked as a best-effort service. We propose an extension to serverless, called real-time serverless that provides an invocation rate guarantee, a service-level objective (SLO) specified by the application, and delivered by the underlying implementation. Real-time serverless allows applications to guarantee real-time performance. We study real-time serverless behavior analytically and empirically to characterize its ability to support bursty, real-time cloud and edge applications efficiently. Finally, we use a case study, traffic monitoring, to illustrate the use and benefits of real-time serverless, on our prototype implementation.