Eman Ramadan

5G aims to offer not only significantly higher throughput than previous generations of cellular networks, but also promises millisecond (ms) and sub-millisecond (ultra-)low latency support at the 5G physical (PHY) layer for future applications. While prior measurement studies have confirmed that commercial 5G deployments can achieve up to several Gigabits per second (Gbps) throughput (especially with the mmWave 5G radio), are they able to deliver on the (sub) millisecond latency promise? With this question in mind, we conducted to our knowledge the first in-depth measurement study of commercial 5G mmWave PHY latency using detailed physical channel events and messages. Through carefully designed experiments and data analytics, we dissect various factors that influence 5G PHY latency of both downlink and uplink data transmissions, and explore their impacts on end-to-end delay. We find that while in the best cases, the 5G (mmWave) PHY-layer is capable of delivering ms/sub-ms latency (with a minimum of 0.09 ms for downlink and 0.76 ms for uplink), these happen rarely. A variety of factors such as channel conditions, re-transmissions, physical layer control and scheduling mechanisms, mobility, and application (edge) server placement can all contribute to increased 5G PHY latency (and thus end-to-end (E2E) delay). Our study provides insights to 5G vendors, carriers as well as application developers/content providers on how to better optimize or mitigate these factors for improved 5G latency performance.

Performance of caching algorithms not only determines the quality of experience for users, but also affects the operating and capital expenditures for cloud service providers. Today's production systems rely on heuristics such as LRU (least recently used) and its variants, which work well for certain types of workloads, and cannot effectively cope with diverse and time-varying workload characteristics. While learning-based caching algorithms have been proposed to deal with these challenges, they still impose assumptions about workload characteristics and often suffer poor generalizability. In this paper, we propose Raven, a general learning-based caching framework that leverages the insights from the offline optimal Belady algorithm for both in-memory and content caching. Raven learns the distributions of objects' next-request arrival times without any prior assumptions by employing Mixture Density Network (MDN)-based universal distribution estimation. It utilizes the estimated distributions to compute the probability of an object that arrives farthest than any other objects in the cache and evicts the one with the largest such probability, regulated by the sizes of objects if appropriate. Raven (probabilistically) approximates Belady by explicitly accounting for the stochastic, time-varying, and non-stationary nature of object arrival processes. Evaluation results on production workloads demonstrate that, compared with the best existing caching algorithms, Raven improves the object hit ratio and byte hit ratio by up to 7.3% and 7.1%, respectively, reduces the average access latency by up to 17.9% and the traffic to the origin servers by up to 18.8%.

As we increasingly depend on networked services, ensuring resiliency of networks against network failures and providing bounded latency to applications become imperative. Adding ample redundancy in the network substrate alone is not sufficient; resilient routing mechanisms that can effectively take advantage of such topological diversity also play a critical role. In this paper, we present Taproot, a resilient diversity routing algorithmthat ensures bounded latencyfor packet delivery under failures by leveraging a preorder Routing structure with precomputed routing rules. Leveraging the centralizedcontrol plane and programmable match-actionrules in the data plane, we describe how Taproot can be realized in SDN networks. We implement Taproot in OVS and conduct extensive simulations and experiments to demonstrate its superior performance over existing solutions. Our results show that by tuning the latency allowance upon failure, Taproot reduces/eliminates the number of disconnected src-dst pairs even under 10 link failures. Finally, as a use case, we illustrate the impact of control channel failures on SDN data plane/application performance, and employ Taproot to provide a "hardened" SDN control network with bounded latency against failures. Our results show that Taproot immediately detects the failure and re-routes the control messages to a different path avoiding failed links/nodes. Hence, the control channel is maintained without interruption, or involvement from the controller, and the throughput was not affected.

Recent measurement studies show that commercial mmWave 5G can indeed offer ultra-high bandwidth (up to 2 Gbps), capable of supporting bandwidth-intensive applications such as ultra-HD (UHD) 4K/8K and volumetric video streaming on mobile devices. However, mmWave 5G also exhibits highly variable throughput performance and incurs frequent handoffs (e.g., between 5G and 4G), due to its directional nature, signal blockage and other environmental factors, especially when the device is mobile. All these issues make it difficult for applications to achieve high Quality of Experience (QoE). In this paper, we advance several new mechanisms to tackle the challenges facing UHD video streaming applications over 5G networks, thereby making them {\em 5G-aware}. We argue for the need to employ machine learning (ML) for effective throughput prediction to aid applications in intelligent bitrate adaptation. Furthermore, we advocate {\em adaptive content bursting}, and {\em dynamic radio (band) switching} to allow the 5G radio network to fully utilize the available radio resources under good channel/beam conditions, whereas dynamically switched radio channels/bands (e.g., from 5G high-band to low-band, or 5G to 4G) to maintain session connectivity and ensure a minimal bitrate. We conduct initial evaluation using real-world 5G throughput measurement traces. Our results show these mechanisms can help minimize, if not completely eliminate, video stalls, despite wildly varying 5G throughput.

5G offers a slew of new features and capabilities to support a whole gamut of new applications. On the other hand, 5G new radio (NR), especially, high-band mmWave radio, also poses new challenges, as shown by recent measurement studies of commercial 5G services. In order to effectively support new classes of application such as extra low-latency and/or high-bandwidth applications, we argue that truly cross-layer network-application integration that exposes application semantics to enable 5G and beyond 5G (B5G) networks to make intelligent decisions, e.g., for dynamic radio resource allocation, is needed. Unfortunately the existing 5G flow-based framework is inadequate to support such cross-layer integration. We therefore advocate a software-defined, fine-grained QoS framework. We use ultra-high resolution (UHR) volumetric video streaming as a use case and conduct very preliminary experiments to demonstrate the potential benefits of the proposed framework. This position paper serves as a strawman to call for new intelligent architectural designs for B5G networks and next-generation wireless systems.

The emerging 5G services offer numerous new opportunities for networked applications. In this study, we seek to answer two key questions: i) is the throughput of mmWave 5G predictable, and ii) can we build "good" machine learning models for 5G throughput prediction? To this end, we conduct a measurement study of commercial mmWave 5G services in a major U.S. city, focusing on the throughput as perceived by applications running on user equipment (UE). Through extensive experiments and statistical analysis, we identify key UE-side factors that affect 5G performance and quantify to what extent the 5G throughput can be predicted. We then propose Lumos5G - a composable machine learning (ML) framework that judiciously considers features and their combinations, and apply state-of-the-art ML techniques for making context-aware 5G throughput predictions. We demonstrate that our framework is able to achieve 1.37x to 4.84x reduction in prediction error compared to existing models. Our work can be viewed as a feasibility study for building what we envisage as a dynamic 5G throughput map (akin to Google traffic map). We believe this approach provides opportunities and challenges in building future 5G-aware apps.

We conduct to our knowledge a first measurement study of commercial 5G performance on smartphones by closely examining 5G networks of three carriers (two mmWave carriers, one mid-band carrier) in three U.S. cities. We conduct extensive field tests on 5G performance in diverse urban environments. We systematically analyze the handoff mechanisms in 5G and their impact on network performance. We explore the feasibility of using location and possibly other environmental information to predict the network performance. We also study the app performance (web browsing and HTTP download) over 5G. Our study consumes more than 15 TB of cellular data. Conducted when 5G just made its debut, it provides a “baseline” for studying how 5G performance evolves, and identifies key research directions on improving 5G users’ experience in a cross-layer manner. We have released the data collected from our study (referred to as 5Gophers) at https://fivegophers.umn.edu/www20.

The emergence of information-centric network (ICN) architectures has attracted a flurry of renewed research interest in caching policies and their performance analysis. One important feature ICNs offer that is distinct from classical computer caches is a distributed network of caches, namely, a cache network which poses additional challenges both in terms of practical cache management issues and performance analysis. Much attention of the research community has focused on performance analysis of cache networks under various caching policies. However, the issue of how to evaluate and compare caching policies for cache networks has not been adequately addressed. In this paper, we propose a novel and general framework for evaluating caching policies in a hierarchical network of caches. We introduce the notion of a hit probability/rate matrix, and employ a generalized notion of majorization as the basic tool for evaluating caching policies for various performance metrics. We discuss how the framework can be applied to existing caching policies, and conduct an extensive simulation-based evaluation to demonstrate the utility and accuracy of our framework.

In this paper, we present Deepcache a novel Framework for content caching, which can significantly boost cache performance. Our Framework is based on powerful deep recurrent neural network models. It comprises of two main components: i) Object Characteristics Predictor, which builds upon deep LSTM Encoder-Decoder model to predict the future characteristics of an object (such as object popularity) - to the best of our knowledge, we are the first to propose LSTM Encoder-Decoder model for content caching; ii) a caching policy component, which accounts for predicted information of objects to make smart caching decisions. In our thorough experiments, we show that applying Deepcache Framework to existing cache policies, such as LRU and k-LRU, significantly boosts the number of cache hits.

In this paper, we develop an optimization decomposition framework for cache management under “BIG” cache abstraction which fully utilizes the cache resources in a cache network. We assign a utility function to each content, and formulate a joint optimization problem to maximize the overall utility of a cache network. We show that this global network utility maximization problem can be decomposed into two sub-problems, the cache allotment problem and object placement problem, which can be solved separately and iteratively. This decoupling enables us to separately optimize the performance objectives from the perspectives of content providers, cache network operators, and users. We provide exact solution to the object placement problem with Poisson and Pareto request interarrival distributions. We also devise a primal-dual algorithm for online content management. We conduct extensive numerical analysis and simulations to evaluate the performance of our optimization decomposition framework, and study the impact of various key factors such as hazard rate functions of the request interarrival distributions and object popularities. We show that our optimization decomposition framework outperform existing heuristic methods.

In this paper, we present DEEPCACHE a novel Framework for content caching, which can significantly boost cache performance. Our Framework is based on powerful deep recurrent neural network models. It comprises of two main components: i) Object Characteristics Predictor, which builds upon deep LSTM Encoder-Decoder model to predict the future characteristics of an object (such as object popularity) -- to the best of our knowledge, we are the first to propose LSTM Encoder-Decoder model for content caching; ii) a caching policy component, which accounts for predicted information of objects to make smart caching decisions. In our thorough experiments, we show that applying DEEPCACHE Framework to existing cache policies, such as LRU and k-LRU, significantly boosts the number of cache hits.

Much attention of the research community has focused on performance analysis of cache networks under various caching policies. However, the issue of how to evaluate and compare caching policies for cache networks has not been adequately addressed. In this paper, we propose a novel and general framework for evaluating caching policies in a hierarchical network of caches. We introduce the notion of a hit probability/rate matrix, and employ a generalized notion of majorization as the basic tool for evaluating caching policies for various performance metrics. We discuss how the framework can be applied to existing caching policies, and conduct extensive simulation-based evaluation to demonstrate the utility and accuracy of our framework.

Building upon the results of recent works on understanding large-scale content distribution systems, we revisit CONIA, a Content-provider Oriented Namespace Independent Architecture for content delivery. The key idea of CONIA is to let any willing ISP or third party to participate as a content distribution network (CDN). In this paper, we propose a first step in the direction of an information-centric network-based open content distribution system (OpenCDN), that allows for better scalability, flexibility, and performance. In particular, we concentrate on the functions of the content store and routing elements (CSRs) that form the network substrate. We propose an actor-model driven programming model and a runtime system, which together we refer to as the OpenCDN platform. Using OpenCDN, content providers will have full control over building and managing the basic building blocks for the functionality of CSRs, and the flexibility on which content to cache, when to cache, and how to satisfy user requests.

In SDN, the logically centralized control plane ("network OS") is often realized via multiple SDN controllers for scalability and reliability. ONOS is such an example, where it employs Raft -- a new consensus protocol developed recently -- for state replication and consistency among the distributed SDN controllers. The reliance of network OS on consensus protocols to maintain consistent network state introduces an intricate inter-dependency between the network OS and the network under its control, thereby creating new kinds of fault scenarios or instabilities. In this paper, we use Raft to illustrate the problems that this inter-dependency may introduce in the design of distributed SDN controllers and discuss possible solutions to circumvent these issues.

In this paper, we advocate the notion of "BIG" cache as an innovative abstraction for effectively utilizing the distributed storage and processing capacities of all servers in a cache network. The "BIG" cache abstraction is proposed to partly address the problem of (cascade) thrashing in a hierarchical network of cache servers, where it has been known that cache resources at intermediate servers are poorly utilized, especially under classical cache replacement policies such as LRU. We lay out the advantages of "BIG" cache abstraction and make a strong case both from a theoretical standpoint as well as through simulation analysis. We also develop the dCLIMB cache algorithm to minimize the overheads of moving objects across distributed cache boundaries and present a simple yet effective heuristic for addressing the cache allotment problem in the design of "BIG" cache abstraction.

In this paper, we propose and advocate a new routing paradigm -- dubbed routing via preorders -- which circumvents the limitations of conventional path-based routing schemes to effectively take advantage of topological diversity inherent in a network with rich topology for adaptive resilient routing, while at the same time meeting the quality-of-service requirements (e.g., latency) of applications or flows. We show how routing via preorders can be realized in SDN networks using the "match-action" data plane abstraction, with a preliminary implementation and evaluation of it in Mininet.

We propose and present CONIA, a novel content (provider)-oriented, namespace-independent architecture for multimedia information delivery. CONIA is designed specifically to account for the diversity and complexity of multimedia content, and to recognize the prominent roles of content providers (CPs) in the network economics of content delivery. In this paper, we provide an overview of the content delivery architecture of CONIA and outline the basic functions of its key components. Using several use cases, we illustrate the flexibility of CONIA in allowing for CPs to employ various control policies to dynamically handle user demands and meet users' quality-of-experience expectations.