As modern server GPUs are increasingly power intensive, better power management mechanisms can significantly reduce the power consumption, capital costs, and carbon emissions in large cloud datacenters. This paper uses diverse datacenter workloads to study the power management capabilities of modern GPUs. We find that current GPU management mechanisms have limited compatibility and monitoring support under cloud virtualization. They have sub-optimal, imprecise, and non-intuitive implementations of Dynamic Voltage and Frequency Scaling (DVFS) and power capping. Consequently, efficient GPU power management is not widely deployed in clouds today. To address these limitations, we make actionable recommendations for GPU vendors and researchers.
Data center network operators often need accurate estimates of aggregate network performance. Unfortunately, existing methods for estimating aggregate network statistics are either inaccurate or too slow to be practical at the data center scale. In this paper, we develop and evaluate a scale-free, fast, and accurate model for estimating data center network tail latency performance for a given workload, topology, and network configuration. First, we show that path-level simulations---simulations of traffic that intersects a given path---produce almost the same aggregate statistics as full network-wide packet-level simulations. We use a simple and fast flow-level fluid simulation in a novel way to capture and summarize essential elements of the path workload, including the effect of cross-traffic on flows on that path. We use this coarse simulation as input to a machine-learning model to predict path-level behavior, and run it on a sample of paths to produce accurate network-wide estimates. Our model generalizes over the choice of congestion control (CC) protocol, CC protocol parameters, and routing. Relative to Parsimon, a state-of-the-art system for rapidly estimating aggregate network tail latency, our approach is significantly faster (5.7×), more accurate (45.9% less error), and more robust.
Climate change is a pressing threat to planetary well-being that can be addressed only by rapid near-term actions across all sectors. Yet, the cloud computing sector, with its increasingly large carbon footprint, has initiated only modest efforts to reduce emissions to date; its main approach today relies on cloud providers sourcing renewable energy from a limited global pool of options. We investigate how to accelerate cloud computing's efforts. Our approach tackles carbon reduction from a software standpoint by gradually integrating carbon awareness into the cloud abstraction. Specifically, we identify key bottlenecks to software-driven cloud carbon reduction, including (1) the lack of visibility and disaggregated control between cloud providers and users over infrastructure and applications, (2) the immense overhead presently incurred by application developers to implement carbon-aware application optimizations, and (3) the increasing complexity of carbon-aware resource management due to renewable energy variability and growing hardware heterogeneity. To overcome these barriers, we propose an agile approach that federates the responsibility and tools to achieve carbon awareness across different cloud stakeholders. As a key first step, we advocate leveraging the role of application operators in managing large-scale cloud deployments and integrating carbon efficiency metrics into their cloud usage workflow. We discuss various techniques to help operators reduce carbon emissions, such as carbon budgets, service-level visibility into emissions, and configurable-yet-centralized resource management optimizations.
Smartphone owners often need to run security-critical programs on the same device as other untrusted and potentially malicious programs. This requires users to trust hardware and system software to correctly sandbox malicious programs, trust that is often misplaced. Our goal is to minimize the number and complexity of hardware and software components that a smartphone owner needs to trust. We present a split-trust hardware design composed of statically-partitioned, physically-isolated trust domains. We introduce a few simple, formally-verified hardware components to enable a program to gain provably exclusive and simultaneous access to both computation and I/O on a temporary basis. To manage this hardware, we present OctopOS, an OS composed of mutually distrustful subsystems. We present a prototype of this machine (hardware and OS) on a CPU-FPGA board and show that it incurs a small hardware cost compared to modern smartphone SoCs. For security-critical programs, we show that this machine significantly reduces the required trust compared to mainstream TEEs while achieving usable performance. For normal programs, performance is similar to a legacy machine.
With the rise of microservices, the execution environment of many cloud applications has become a set of virtual machines or containers connected by a flexible and feature-rich virtual network. We argue that the implementation of such virtual networks should be completely application-specific and not layered on top of general-purpose network abstractions from the Internet age. Such layering tends to more than double the latency and CPU usage of applications. We propose application-defined networks in which developers specify network functionality in a high-level language and a controller generates a custom distributed implementation that runs across available hardware and software resources. Experiments with a preliminary prototype suggest that, compared to the state of the art, ADN reduces latency by up to 20x and increases the throughput by up to 6x.