Interest has been growing in powering datacenters (at least partially) with renewable or "green" sources of energy, such as solar wind. However, it is challenging to use these because, unlike the "brown" (carbon-intensive) energy drawn from electrical grid, they are not always available. This means that demand and supply must be matched, if we take full advantage green minimize brown consumption. In this paper, investigate how manage a datacenter's computational workload match supply....

In this paper, we propose GreenSlot, a parallel batch job scheduler for datacenter powered by photovoltaic solar array and the electrical grid (as backup). GreenSlot predicts amount of energy that will be available in near future, schedules workload to maximize green consumption while meeting jobs' deadlines. If must used avoid deadline violations, selects times when it is cheap. Our results production scientific workloads demonstrate Green-Slot can increase up 117% decrease cost 39%,...

As energy-related costs have become a major economical factor for IT infrastructures and data-centers, companies the research community are being challenged to find better more efficient power-aware resource management strategies. There is growing interest in "Green" there still big gap this area be covered.

Several companies have recently announced plans to build "green" datacenters, i.e. datacenters partially or completely powered by renewable energy. These will either generate their own energy draw it directly from an existing nearby plant. Besides reducing carbon footprints, can potentially reduce costs, peak power both. However, certain fuels are intermittent, which requires approaches for tackling the supply variability. One approach is use batteries and/or electrical grid as a backup It...

Function as a Service (FaaS) has been gaining popularity way to deploy computations serverless backends in the cloud. This paradigm shifts complexity of allocating and provisioning resources cloud provider, which provide illusion always-available (i.e., fast function invocations without cold starts) at lowest possible resource cost. Doing so requires provider deeply understand characteristics FaaS workload. Unfortunately, there little no public information on these characteristics. Thus,...

We propose and evaluate a framework for creating running approximation-enabled MapReduce programs. Specifically, we approximation mechanisms that fit naturally into the paradigm, including input data sampling, task dropping, accepting precise user-defined approximate version of code. then show how to leverage statistical theories compute error bounds popular classes programs when approximating with sampling and/or dropping. implement proposed bound estimations in prototype system called...

Cloud federation has been proposed as a new paradigm that allows providers to avoid the limitation of owning only restricted amount resources, which forces them reject customers when they have not enough local resources fulfill their customers' requirements. Federation provider dynamically outsource other in response demand variations. It also underused rent part providers. Both things could make get more profit used adequately. This requires clear understanding potential each decision,...

Modern resource management frameworks for large-scale analytics leave unresolved the problematic tension between high cluster utilization and job's performance predictability--respectively coveted by operators users. We address this in Morpheus, a new system that: 1) codifies implicit user expectations as explicit Service Level Objectives (SLOs), inferred from historical data, 2) enforces SLOs using novel scheduling techniques that isolate jobs sharing-induced variability, 3) mitigates...

Popular Internet services are hosted by multiple geographically distributed data centers. The location of the centers has a direct impact on services' response times, capital and operational costs, (indirect) carbon dioxide emissions. Selecting involves many important considerations, including its proximity to population centers, power plants, network backbones, source electricity in region, electricity, land, water prices at location, average temperatures location. As there can be potential...

Serverless computing is becoming increasingly popular due to its ease of programming, fast elasticity, and fine-grained billing. However, the serverless provider still needs provision, manage, pay IaaS for virtual machines (VMs) hosting platform. This ties cost platform underlying VMs. One way significantly reduce use spare resources, which cloud providers rent at a massive discount. Harvest VMs offer such cheap resources: they grow shrink harvest all unallocated CPU cores in their host...

Function-as-a-Service (FaaS) has become an increasingly popular way for users to deploy their applications without the burden of managing underlying infrastructure. However, existing FaaS platforms rely on remote storage maintain state, limiting set that can be run efficiently. Recent caching work tried address this problem, but fallen short: it disregards widely different characteristics applications, does not scale cache based data access patterns, or requires changes applications. To...

Recent innovation in large language models (LLMs), and their myriad use cases have rapidly driven up the compute demand for datacenter GPUs. Several cloud providers other enterprises plan to substantially grow capacity support these new workloads. A key bottleneck resource datacenters is power, which LLMs are quickly saturating due increasing model sizes.

The reduction of energy consumption in large-scale datacenters is being accomplished through an extensive use virtualization, which enables the consolidation multiple workloads a smaller number machines. Nevertheless, virtualization also incurs some additional overheads (e.g. virtual machine creation and migration) that can influence what best consolidated configuration, thus, they must be taken into account. In this paper, we present dynamic job scheduling policy for power-aware resource...

An effective way to increase utilization and reduce costs in datacenters is co-locate their latency-critical services batch workloads. In this paper, we describe systems that harvest spare compute cycles storage space for co-location purposes. The main challenge minimizing the performance impact on services, while accounting management patterns. To overcome challenge, propose techniques giving priority over resources, leveraging historical information about them. Based information, schedule...

Computing capacity has always been on the upward climb due to constant technological improvements in semiconductor manufacturing and packaging industry. This growth computing capability is usually accompanied by a steep rise heat flux density associated with electronic component (CPUs or GPUs, for example). High-performance (HPC) data centers often employ several of these high-performance devices crunching their enormous artificial intelligence (AI) scientific workloads. State-of-the-art air...

Function-as-a-Service (FaaS) serverless computing enables a simple programming model with almost unbounded elasticity. Unfortunately, current FaaS platforms achieve this flexibility at the cost of lower performance for data-intensive applications compared to serverful deployment. The ability have computation close data is key missing feature. We introduce Palette load balancing, which offers mechanism express locality platform, through hints we term "colors". maintains nature service - users...

On-site renewable energy has the potential to reduce datacenters' carbon footprint and power costs. The authors built Parasol, a solar-powered datacenter, GreenSwitch, system for scheduling workloads, explore this in controlled research setting.

Cloud providers typically use air-based solutions for cooling servers in datacenters. However, increasing transistor counts and the end of Dennard scaling will result chips with thermal design power that exceeds capabilities air near future. Consequently, have started to explore liquid (e.g., cold plates, immersion cooling) most power-hungry workloads. By keeping cooler, these new enable operate server components beyond normal frequency range (i.e., overclocking them) all time. Still, must...

Crash and omission failures are common in service providers: a disk can break down or link fail anytime. In addition, the probability of node failure increases with number nodes. Apart from reducing provider's computation power jeopardizing fulfillment his contracts, this also lead to time wasting when crash occurs before finishing task execution. order avoid problem, efficient checkpoint infrastructures required, especially virtualized environments where these must deal huge virtual machine...

Demand for clean products and services is increasing as society becoming increasingly aware of climate change. In response, many enterprises are setting explicit sustainability goals implementing initiatives to reduce carbon emissions. Quantification disclosure such have become important marketing tools. As individuals shift their workloads the cloud, this drive toward quantification will lead demand quantifiable green cloud services. Thus, we argue that providers should offer a new class...