Due to limited downlink (satellite-to-ground) capacity, over 90% of the images captured by the earth-observation satellites are not downloaded to the ground. To overcome the downlink limitation, we present Earth+, a new on-board satellite imagery compression system that identifies and downloads only changed areas in each image compared to latest on-board reference images of the same location. The key of Earth+ is that it obtains latest on-board reference images by letting the ground stations upload images recently captured by all satellites in the constellation. To our best knowledge, Earth+ is the first system that leverages images across an entire satellite constellation to enable more images to be downloaded to the ground (by better satellite imagery compression). Our evaluation shows that to download images of the same area, Earth+ can reduce the downlink usage by 3.3× compared to state-of-the-art on-board image compression techniques without sacrificing imagery quality or using more resources (downlink, computation or storage).

Modular data centers (MDCs) that can be placed right at the energy farms and powered mostly by renewable energy, is a flexible and effective approach to lowering the carbon footprint of data centers. However, the main challenge of using renewable energy is the high variability of power produced, which implies large volatility in powering computing resources at MDCs, and degraded application performance due to the task evictions and migrations. This causes challenges for platform operators to decide the MDC deployment.

Decreasing costs of deploying space vehicles to low-Earth orbit have fostered an emergence of large constellations of satellites. However, high satellite velocities, large image data quantities, and brief ground station contacts create a data downlink challenge. Orbital edge computing (OEC), which filters data at the space edge, addresses this downlink bottleneck but shifts the challenge to the inelastic computational capabilities onboard satellites. In this work, we present Kodan: an OEC system that maximizes the utility of saturated satellite downlinks while mitigating the computational bottleneck. Kodan consists of two phases. A one-time transformation step uses a reference implementation of a satellite data analysis application, along with a representative dataset, to produce specialized ML models targeted for deployment to the space edge. After deployment to a target satellite, a runtime system dynamically selects the best specialized models for each data sample to maximize valuable data downlinked within the constraints of the computational bottleneck. By intelligently filtering low-value data and prioritizing high-value data for transmit via the saturated downlink, Kodan increases the data value density between 89 and 97 percent.

Modern developers rely on container-orchestration frameworks like Kubernetes to deploy and manage hybrid workloads that span the edge and cloud. When network conditions between the edge and cloud change unexpectedly, a workload must adapt its internal behavior. Unfortunately, container-orchestration frameworks do not offer an easy way to express, deploy, and manage adaptation strategies. As a result, fine-tuning or modifying a workload's adaptive behavior can require modifying containers built from large, complex codebases that may be maintained by separate development teams. This paper presents BumbleBee, a lightweight extension for container-orchestration frameworks that separates the concerns of application logic and adaptation logic. BumbleBee provides a simple in-network programming abstraction for making decisions about network data using application semantics. Experiments with a BumbleBee prototype show that edge ML-workloads can adapt to network variability and survive disconnections, edge stream-processing workloads can improve benchmark results between 37.8% and$\boldsymbol{23\mathrm{x}}$, and HLS video-streaming can reduce stalled playback by 77%.

Smart environments (homes, factories, hospitals, buildings) contain an increasing number of IoT devices, making them complex to manage. Today, in smart homes when users or triggers initiate routines (i.e., a sequence of commands), concurrent routines and device failures can cause incongruent outcomes. We describe SafeHome, a system that provides notions of atomicity and serial equivalence for smart homes. Due to the human-facing nature of smart homes, SafeHome offers a spectrum of visibility models which trade off between responsiveness vs. isolation of the smart home. We implemented SafeHome and performed workload-driven experiments. We find that a weak visibility model, called eventual visibility, is almost as fast as today's status quo (up to 23% slower) and yet guarantees serially-equivalent end states.

Renewable energy is becoming an important power source for data centers, especially with the zero-carbon waste pledges made by big cloud providers. However, one of the main challenges of renewable energy sources is the high variability of power produced. Traditional approaches such as batteries or transmitting to the grid fall short on scale, overhead, or "green-ness". We propose Virtual Battery: instead of adapting the availability of power to match the computation demand we shift computational demand to meet the availability of power. Virtual batteries shift demand by requiring applications to either be flexible and delay-tolerant or proactively migrating to where power is (going to be) available. We show that using multiple virtual battery sites in combination can meet the needs of modern applications. Moreover, we show how an intelligent network and power aware co-scheduler can not only provide availability despite variability but also help mitigate migration related network overhead by over 30% in total and 4.2x at peak.

Analytics with three-dimensional imagery from drones are driving the next generation of remote monitoring applications. Today, there is an unmet need in providing such analytics in an interactive manner, especially over weak Internet connections, to quickly diagnose and solve problems in the commercial industry space of monitoring assets using drones in remote parts of the world. Existing mechanisms either compromise on the quality of insights by not building 3D images and analyze individual 2D images in isolation, or spend tens of minutes building a 3D image before obtaining and uploading insights. We present Visage, a system that accelerates 3D image analytics by identifying smaller parts of the data that can actually benefit from 3D analytics and prioritizing building, and uploading the localized 3D images for those parts. To achieve this, Visage uses a graph to represent raw 2D images and their relative content overlap, and then identifies the various subgraphs using application knowledge that are good candidates for localized 3D image based insights. We evaluate Visage using data from multiple real deployments and show that it can reduce analytics-latency by up to four orders of magnitude.

The world is witnessing an unprecedented increase in camera deployment. The USA and UK, for instance, have one camera for every 8 people. Video analytics from these cameras are becoming more and more pervasive, exerting important functions on a wide range of verticals including manufacturing, transportation, and retails. While vision techniques have seen considerable advancement, they have come at the expense of compute and network cost.

Edge computing caters to a wide range of use cases from latency sensitive to bandwidth constrained applications. However, the exact specifications of the edge that give the most benefit for each type of application are still unclear. We investigate the concrete conditions when the edge is feasible, i.e., when users observe performance gains from the edge while costs remain low for the providers, for an application that requires both low latency and high bandwidth: video analytics.

Distributed stream processing systems need to support stateful processing, recover quickly from failures to resume such processing, and reprocess an entire data stream quickly. We present Apache Samza, a distributed system for stateful and fault-tolerant stream processing. Samza utilizes a partitioned local state along with a low-overhead background changelog mechanism, allowing it to scale to massive state sizes (hundreds of TB) per application. Recovery from failures is sped up by re-scheduling based on Host Affinity. In addition to processing infinite streams of events, Samza supports processing a finite dataset as a stream, from either a streaming source (e.g., Kafka), a database snapshot (e.g., Databus), or a file system (e.g. HDFS), without having to change the application code (unlike the popular Lambda-based architectures which necessitate maintenance of separate code bases for batch and stream path processing). Samza is currently in use at LinkedIn by hundreds of production applications with more than 10, 000 containers. Samza is an open-source Apache project adopted by many top-tier companies (e.g., LinkedIn, Uber, Netflix, TripAdvisor, etc.). Our experiments show that Samza: a) handles state efficiently, improving latency and throughput by more than 100X compared to using a remote storage; b) provides recovery time independent of state size; c) scales performance linearly with number of containers; and d) supports reprocessing of the data stream quickly and with minimal interference on real-time traffic.

Many in the networking community believe that Software-Defined Networking, in which entire networks are managed centrally, has the potential to revolutionize the field. However, SDN faces several challenges that have prevented its wide-spread adoption. Current SDN technologies, such as OpenFlow, provide powerful and flexible APIs, but can be unreasonably complex for implementing nontrivial network control logic. The generality offered by these low-level abstractions impose no structure on the network, requiring programmers to herd switches themselves, with little guidance. Many researchers argue that SDNs must adopt more structured models, such as Fabric, with an intelligent edge and a fast but simple label-switched core. Our work draws heavily from these ideas.

With the tremendous popularity gained by container technology, many applications are being containerized: splitting into numerous containers connected by networks. However, current container networking solutions have either bad performance or poor portability, which undermines the advantages of containerization. In this paper, we propose FreeFlow, a container networking solution which achieves both high performance and good portability. FreeFlow is designed according to the observation that strict isolations are unnecessary among containers trusting each other, and it can significantly boost the communication quality of containers by compromising isolation a little bit. Specifically, we enable containers on the same physical machine to communicate via shared-memory and the ones on different physical machines communicate via high performance networking options, e.g. RDMA and DPDK. Naively wrapping up all the solutions together will result in poor potability of containers and huge complexity in application development. Instead, FreeFlow leverages a network abstraction which supports all common network APIs and a centralized network orchestrator which decides how to deliver data transparently to applications in the containers.

The infrastructure beneath a worldwide social network has to continually serve billions of variable-sized media objects such as photos, videos, and audio clips. These objects must be stored and served with low latency and high throughput by a system that is geo-distributed, highly scalable, and load-balanced. Existing file systems and object stores face several challenges when serving such large objects. We present Ambry, a production-quality system for storing large immutable data (called blobs). Ambry is designed in a decentralized way and leverages techniques such as logical blob grouping, asynchronous replication, rebalancing mechanisms, zero-cost failure detection, and OS caching. Ambry has been running in LinkedIn's production environment for the past 2 years, serving up to 10K requests per second across more than 400 million users. Our experimental evaluation reveals that Ambry offers high efficiency (utilizing up to 88% of the network bandwidth), low latency (less than 50 ms latency for a 1 MB object), and load balancing (improving imbalance of request rate among disks by 8x-10x).