Showing papers by "Rich Wolski published in 2019"

CSPOT: portable, multi-scale functions-as-a-service for IoT

Abstract: In this paper, we present CSPOT, a distributed runtime system implementing a functions-as-service (FaaS) programming model for the "Internet of Things" (IoT). With FaaS, developers express arbitrary computations as simple functions that are automatically invoked and managed by a cloud platform in response to events. We extend this FaaS model so that it is suitable for use in all tiers of scale for IoT - sensors, edge devices, and cloud - to facilitate robust, portable, and low-latency IoT application development and deployment. To enable this, we combine the use of Linux containers and namespaces for isolation and portability, an append-only object store for robust persistence, and a causal event log for triggering functions and tracking event dependencies. We present the design and implementation of CSPOT, detail its abstractions and APIs, and overview examples of its use. We empirically evaluate the performance of CSPOT using different devices and applications and find that it implements function invocation with significantly lower latency than other FaaS offerings, while providing portability across tiers and similar data durability characteristics.

Deconstructing the 2017 Changes to AWS Spot Market Pricing

Abstract: The Amazon Web Services spot market sells excess computing capacity at a reduced price and with reduced reliability guarantees. The low cost nature of the spot market has led to widespread adoption in industry and science. However, one of the challenges with using the spot market is that it is intentionally opaque and thus users have little understanding of the underlying dynamics. In late 2017, the mechanisms underlying the spot market were significantly altered-no longer are bid prices used to clear capacity and as a result the pricing is much less volatile. In this paper, we revisit prior work with the aim to analyze the differences in market dynamics between the pre-change and post-change spot instance market. We then use these analyses to highlight possible properties of the current and previous pricing algorithms, including artificial manipulation, dynamic algorithm adjustment, and persistent trends in market supply, demand, and pricing.

Data Repair for Distributed, Event-based IoT Applications

Abstract: Motivated by the growth of Internet of Things (IoT) technologies and the volumes and velocity of data that they can and will produce, we investigate automated data repair for event-driven, IoT applications. IoT devices are heterogeneous in their hardware architectures, software, size, cost, capacity, network capabilities, power requirements, etc. They must execute in a wide range of operating environments where failures and degradations of service due to hardware malfunction, software bugs, network partitions, etc. cannot be immediately remediated. Further, many of these failure modes cause corruption in the data that these devices produce and in the computations "downstream" that depend on this data.To "repair" corrupted data from its origin through its computational dependencies in a distributed IoT setting, we explore SANS-SOUCI--a system for automatically tracking causal data dependencies and re-initiating dependent computations in event-driven IoT deployment frameworks. SANS-SOUCI presupposes an event-driven programming model based on cloud functions, which we extend for portable execution across IoT tiers (device, edge, cloud). We add fast, persistent, append-only storage and versioning for efficient data robustness and durability. SANS-SOUCI records events and their causal dependencies using a distributed event log and repairs applications dynamically, across tiers via replay. We evaluate SANS-SOUCI using a portable, open source, distributed IoT platform, example applications, and microbenchmarks. We find that SANS-SOUCI is able to perform repair for both software (function) and sensor produced data corruption with very low overhead.

Analyzing AWS Spot Instance Pricing

Abstract: Many cloud computing vendors offer a preemptible class of service for rented virtual machines. In November 2017, Amazon.com changed the pricing mechanism for its preemptible "spot instances" so that prices would change more "smoothly." This paper analyzes the effect of this change on spot instance prices. It examines the prices immediately before and after the mechanism change to determine the extent to which prices themselves changed. It then compares the 90-day period immediately after the change in mechanism to the next 90-day period. Finally, it compares the two most recent 90-day periods (ending on October 15, 2018). Our results indicate that in addition to smoothing prices, the mechanism change introduced generally higher prices which is a trend that continues.

Seneca: Fast and Low Cost Hyperparameter Search for Machine Learning Models

Abstract: The goal of our work is to simplify and expedite the construction and evaluation of machine learning models using autoscaled cloud computing resources. To enable this, we develop an open source system called Seneca, which leverages the serverless programming model and its implementation in Amazon Web Services (AWS) Lambda. Seneca takes a machine learning application, dataset, and a list of possible hyperparameter options as input and automatically constructs an AWS Lambda function. The function ingresses and splits the input dataset into training and testing subsets and constructs, tests, and evaluates (i.e. scores) a machine learning model for a given set of hyperparameter values. Seneca concurrently invokes functions for all combinations of the hyperparameters specified. It then returns the configuration (or model) that results in the best score to the user. In this paper, we overview the design and implementation of Seneca, and empirically evaluate its performance for a popular classification application.

Improving the Accuracy of Outdoor Temperature Prediction by IoT Devices

Abstract: In this paper, we investigate new methods for improving the accuracy of outdoor temperature prediction using small, low-cost, single board computers (SBCs) used in Internet-of-Things (IoT) deployments. Predicting temperature without dedicated temperature sensors frees up space on these systems for other sensors and reduces the cost of microclimate sensing (e.g. as used in IoT-based, agricultural applications). Our approach uses multiple linear regression and combines measurements of on-board processor temperature from multiple SBCs with remote weather stations. In addition, it accounts for SBC computational load through the use of smoothing techniques that filter out noise in the measurement time series. We empirically evaluate our approach using multiple IoT deployment scenarios, compare it against prior work, and find that it reduces prediction error significantly for these scenarios.

Mandrake: Implementing Durability for Edge Clouds

Abstract: In this paper we present Mandrake, a software infrastructure for edge clouds (private clouds located at the network edge), designed to provide reliable, "lights out" unattended operation and application hosting in IoT deployments. Mandrake both implements reliable private cloud operation in restricted resource environments and data durability features that hosted applications can leverage. We describe leveraging Mandrake for hosting Hadoop applications at the edge. Our empirical evaluation shows that Mandrake is able to ensure Hadoop's data durability guarantees efficiently in the presence of relatively frequent failures even when resources are scarce.

A scalable system for executing and scoring K-means clustering techniques and its impact on applications in agriculture

Abstract: We present Centaurus - a scalable, open source, clustering service for K-means clustering of correlated, multidimensional data. Centaurus provides users with automatic deployment via public or private cloud resources, model selection (using Bayesian information criterion), and data visualisation. We apply Centaurus to a real-world, agricultural analytics application and compare its results to the industry standard clustering approach. The application uses soil electrical conductivity (EC) measurements, GPS coordinates, and elevation data from a field to produce a 'map' of differing soil zones (so that management can be specialised for each). We use Centaurus and these datasets to empirically evaluate the impact of considering multiple K-means variants and large numbers of experiments. We show that Centaurus yields more consistent and useful clusterings than the competitive approach for use in zone-based soil decision-support applications where a 'hard' decision is required.

Towards distributed, fair, deadline-driven resource allocation for cloudlets

Abstract: In this paper we present our vision for a two-level, distributed resource allocator that preserves fairness and satisfies deadlines of low latency workloads in a multi-cloudlet environment with offloading support. We analyze the opportunities and challenges that offloading and the multi-cloud environment impose and we suggest the changes required to a fair-preserving and deadline-driven resource allocator originally designed for resource-constrained environments.

Devices-as-Services: Rethinking Scalable Service Architectures for the Internet of Things

Abstract: We investigate a new distributed services model and architecture for Internet of Things (IoT) applications. In particular, we observe that devices at the edge of the network, although resource constrained, are increasingly capable – performing actions (e.g. data analytics, decision support, actuation, control, etc.) in addition to event telemetry. Thus, such devices are better modeled as servers, which applications in the cloud compose for their functionality. We investigate the implications of this “flipped” IoT client-server model, for server discovery, authentication, and resource use. We find that by combining capability-based security with an edge-aware registry, this model can achieve fast response and energy efficiency.