Technology companies have been leading the way to a renewable energy transformation, by investing in sources reduce carbon footprint of their datacenters. In addition helping build new solar and wind farms, make power purchase agreements or offsets, rather than relying on every hour day, day week (24/7). Relying 24/7 is challenging due intermittent nature energy. Inherent variations production causes excess lack supply at different times. To cope with fluctuations generation, multiple...

We consider the problem of allocating a capital budget to solar panels and storage maximize expected revenue in context large-scale farm participating an energy market. This is complex due many factors. To begin with, production stochastic, with high peak-to-average ratio, thus access link typically provisioned at less than peak capacity, leading potential waste curtailment. The use prevents power curtailment, but allocation reduces amount produced. Moreover, devices are imperfect. A owner...

Compute Express Link (CXL) is widely-supported interconnect standard that promises to enable memory disaggregation in data centers. CXL allows for pooling, which can be used create a shared space across multiple servers. However, does not specify how actually build pool. Existing proposals pools are expensive, as they require switches or large multi-headed devices. In this paper, we propose new design cost-effective. We call these designs Octopus topologies. Our uses small devices made...

Resource allocation problems in many computer systems can be formulated as mathematical optimization problems. However, finding exact solutions to these using off-the-shelf solvers is often intractable for large problem sizes with tight SLAs, leading system designers rely on cheap, heuristic algorithms. We observe, however, that are granular: they consist of a number clients and resources, each client requests small fraction the total interchangeably use different resources. For problems, we...

Motivated by the increase in small-scale solar installations used for powering homes and small businesses, we consider design of rule-based strategies operating an energy storage device connected to a self-use generation system minimize payments grid. This problem is inherently challenging, since depend greatly on choice tariff structure forecasts future load. We propose optimization framework finding optimal operation use it evaluate performance existing strategy that modified not...

Building a sustainable datacenter requires coordinated decisions in its design and system management. Existing research work on sustainability often considers the space isolation misses opportunities to minimize carbon footprint through management where is first-class objective. Design such as site selection, renewable energy investment portfolios, provisioning of storage are intertwined with complementary solutions for operation, including various forms demand response carbon-aware workload...

There is a need for accurate analytical models that describe how Lithium-ion battery's state of charge evolves as result charging or discharging operation and can be used in optimization problems. Although 'white box' take into account the details electro-chemical processes highly accurate, they are not typically suitable We propose two represent different tradeoffs between accuracy tractability. validate these with data traces obtained from extensive experiments using commercially-available...

Lithium-ion battery models that estimate their energy content after a series of charge and discharge operations are essential in the optimal design, analysis, operation battery-based systems. We focus on class can be calibrated entirely from batteries manufacturer-provided specifications (spec). Such simple to calibrate therefore widely used practice. The best-known model this category was proposed by Tremblay et al. 2007. This model, however, has several shortcomings, including low fidelity...

Major cloud providers have stated public plans to lower their carbon emissions. Historically, this has meant focusing on emissions from producing the electricity consumed by datacenters. While work and challenges remain avenue, research industry are actively working next step of reducing embedded in servers racks. At a high level, promising direction reduce embodied is avoid new manufacturing, which often requires using existing components, devices, buildings for longer. However, much data...

We study the problem of optimally and simultaneously sizing solar photovoltaic (PV) storage capacity in order to partly or completely offset grid usage. While prior work offers some insights, researchers typically consider only a single approach. In contrast, we use firm theoretical foundation compare contrast approaches based on robust simulation, optimization, stochastic network calculus. evaluate robustness computational complexity these realistic setting provide practical, advice system sizing.

In this paper, we introduce Mayflower, a new distributed filesystem that is co-designed from the ground up to work together with network control plane. addition standard components, Mayflower has flow monitor and manager running alongside software-defined networking controller. This tight coupling controller enables make intelligent replica selection scheduling decisions based on both information. It further perform global optimizations are unavailable conventional filesystems planes. Our...

The return on investment of a battery system is maximized if the control strategy appropriately matched to operating environment (e.g., pricing scheme, electrical load). For residential systems, current practice statically determine policy prior installation; subsequently spends upwards 10 years in dynamic environment. A state-of-the-art model predictive controller (MPC) can adapt changes system, but limited by its high online computational requirements. To better extract value at reasonable...

We study the problem of optimally and simultaneously sizing solar photovoltaic (PV) storage capacity in order to partly or completely offset grid usage. While prior work offers some insights, researchers typically consider only a single approach. In contrast, we use firm theoretical foundation compare contrast approaches based on robust simulation, optimization, stochastic network calculus. evaluate robustness computational complexity these realistic setting provide practical, advice system sizing.

Model-based optimization of energy systems with batteries requires a battery model that is accurate, tractable, and easy to calibrate. Developing such challenging because electrochemical exhibit complex behaviours. In this paper, we propose evaluate family models have different trade-offs between accuracy complexity. We derive our from recently developed model, which accurate calibrate but not tractable. against the commonly-used benchmark tractable using set experiments characterize cycling...

Due to the inherent randomness of both solar power generation and residential electrical load, jointly sizing panel storage capacity meet a given quality-of-service (QoS) constraint is challenging. The challenge greater when there limited representative historical data. We therefore propose generating synthetic load traces, corresponding different realizations underlying stochastic processes. Specifically, we compare effectiveness three generative models: autoregressive moving-average (ARMA)...

Reinforcement learning (RL) is often considered a promising approach for controlling complex building operations. In this context, RL algorithms are typically evaluated using testing framework that simulates To make general claims and avoid overfitting, an method should be on large diverse set of buildings. Unfortunately, due to the complexity creating simulations, none existing frameworks provide more than handful simulated Moreover, each has its own particularities, which makes it...

Specialized accelerators such as GPUs, TPUs, FPGAs, and custom ASICs have been increasingly deployed to train deep learning models. These exhibit heterogeneous performance behavior across model architectures. Existing schedulers for clusters of accelerators, which are used arbitrate these expensive training resources many users, shown how optimize various multi-job, multi-user objectives, like fairness makespan. Unfortunately, existing largely do not consider heterogeneity. In this paper, we...

Increasing the use of renewable energy is considered a viable way reducing carbon intensive power generation. However, grid running on high amounts has to deal with limited controllability and higher volatility sources like wind or solar. In this work, we propose demand side management varying feed-in via plans, i.e. instructions passed large consumers that specify how they should try spread out their over day. We argue separation planning implementation technical measures schedule loads...

Systems performing large data-parallel computations, including online analytical processing (OLAP) systems like Druid and search engines Elasticsearch, are increasingly being used for business-critical real-time applications where providing low query latency is paramount. In this paper, we investigate an underexplored factor in the performance of queries: their parallelism. We find that to minimize tail queries, it critical place data such items accessed by each individual spread across as...

Choosing the number of solar panels and amount storage needed to meet a certain fraction load in microgrid setting is difficult problem that needs balance competing objectives efficiency, robustness, cost. Prior work this area makes unrealistic assumption are be installed on single roof capable supporting all required. In reality, we may need deploy several segments, each limited size, with its own tilt, orientation installation This paper presents an algorithm for sizing context. We...