Serverless computing has emerged as a new cloud computing paradigm, where an application consists of individual functions that can be separately managed and executed. However, the function development environment of all serverless computing frameworks at present is CPU-based. In this paper, we propose to extend the open-sourced KNIX high-performance serverless framework so that it can execute functions on shared GPU cluster resources. We have evaluated the performance impacts on the extended KNIX system by measuring overheads and penalties incurred using different deep learning frameworks.

Communication networks need to be both adaptive and scalable. The last few years have seen an explosive growth of software-defined networking (SDN) and network function virtualization (NFV) to address this need. Both technologies help enable networking software to be decoupled from the hardware, so that software functionality is no longer constrained by the underlying hardware and can evolve independently. Both SDN and NFV aim to advance a software-based approach to networking, where networking functionality is implemented in software modules and executed on a suitable cloud computing platform. Achieving this goal requires the virtualization paradigm used in these services play an important role in the transition to software-based networks. Consequently, the corresponding computing platforms accompanying the virtualization technologies need to provide the required agility, robustness and scalability for the services executed. Serverless Computing has recently emerged as a new paradigm in virtualization and has already significantly changed the economics of offloading computations to the cloud. It is considered as a low latency, resource-efficient, and rapidly deployable alternative to traditional virtualization approaches, such as those based on virtual machines and containers. Serverless Computing provides scalability and cost reduction, without requiring any additional configuration overhead on the part of the developer. In this paper, we explore and survey how Serverless Computing technology can help building adaptive and scalable networks, and show the potential pitfalls of doing so.

Communication networks need to be both adaptive and scalable. The last few years have seen an explosive growth of software-defined networking (SDN) and network function virtualization (NFV) to address this need. Both technologies help enable networking software to be decoupled from the hardware so that software functionality is no longer constrained by the underlying hardware and can evolve independently. Both SDN and NFV aim to advance a software-based approach to networking, where networking functionality is implemented in software modules and executed on a suitable cloud computing platform. Achieving this goal requires the virtualization paradigm used in these services that play an important role in the transition to software-based networks. Consequently, the corresponding computing platforms accompanying the virtualization technologies need to provide the required agility, robustness, and scalability for the services executed. Serverless computing has recently emerged as a new paradigm in virtualization and has already significantly changed the economics of offloading computations to the cloud. It is considered as a low-latency, resource-efficient, and rapidly deployable alternative to traditional virtualization approaches, such as those based on virtual machines and containers. Serverless computing provides scalability and cost reduction, without requiring any additional configuration overhead on the part of the developer. In this paper, we explore and survey how serverless computing technology can help building adaptive and scalable networks and show the potential pitfalls of doing so.

Serverless computing has emerged as a new cloud computing paradigm, where an application consists of individual functions that can be separately managed and executed. However, existing serverless platforms normally isolate and execute functions in separate containers, and do not exploit the interactions among functions for performance. These practices lead to high startup delays for function executions and inefficient resource usage. This paper presents SAND, a new serverless computing system that provides lower latency, better resource efficiency and more elasticity than existing serverless platforms. To achieve these properties, SAND introduces two key techniques: 1) application-level sandboxing, and 2) a hierarchical message bus. We have implemented and deployed a complete SAND system. Our results show that SAND outperforms the state-of-the-art serverless platforms significantly. For example, in a commonly-used image processing application, SAND achieves a 43% speedup compared to Apache OpenWhisk.

Online video presents new challenges to traditional caching with over a thousand-fold increase in number of assets, rapidly changing popularity of assets and much higher throughput requirements. We propose a new hierarchical filtering algorithm for caching online video HiFi. Our algorithm is designed to optimize hit rate, replacement rate and cache throughput. It has an associated implementation complexity comparable to that of LRU. Our results show that, under typical operator conditions, HiFi can increase edge cache byte hit rate by 5%--24% over an LRU policy, but more importantly can increase the RAM or memory byte hit rate by 80% to 200% and reduce the replacement rate by more than 100 times! These two factors combined can dramatically increase throughput for most caches. If SSDs are used for storage, the much lower replacement rate may also allow substitution of lower-cost MLC-based SSDs instead of SLC-based SSDs. We extend previous multi-tier analytical models for LRU caches to caches with filtering. We analytically show how HiFi can approach the performance of an optimal caching policy and how to tune HiFi to reach as close to optimal performance as the traffic conditions allow. We develop a realistic simulation environment for online video using statistics from operator traces. We show that HiFi performs within a few percentage points from the optimal solution which was simulated by Belady's MIN algorithm under typical operator conditions

Online video presents new challenges to traditional caching with over a thousand fold increase in number of assets, rapidly changing popularity of assets and much higher throughput requirements. We propose a new hierarchical filtering algorithm for caching online video-HiFi. Our algorithm is designed to optimize hit-rate, replacement rate and cache throughput. It has an associated implementation complexity comparable to that of LRU. Our results show that under typical operator conditions, HiFi can increase edge cache byte hit-rate by 5-24% over an LRU policy, but more importantly can increase RAM or memory byte hit-rate by 80% to 200% and reduce replacement rate by 90%! These two factors combined can dramatically increase throughput for most caches. If SSDs are used for storage, the much lower replacement rate may also allow substitution of lower cost MLC based SSDs instead of SLC based SSDs. We extend previous multi-tier analytical models for LRU caches to caches with filtering. We develop a realistic simulation environment for online video using statistics from operator traces. We show that HiFi performs within a few percentage points from the optimal solution which was simulated by Belady's MIN algorithm under typical operator conditions.

In this paper, we describe WiLo, a Rate Determination Algorithm (RDA) for HTTP Adaptive Streaming (HAS) applications. We demonstrate that WiLo delivers stable video output and avoids buffer starvation, even in highly dynamic wireless networks where existing algorithms often struggle. We also show WiLo to be effective with very small buffers, making it useful for live streaming applications where delay is a concern.

Video is expected to be the dominant application by traffic volume over mobile networks in the near future. Mobile network operators are deploying video optimization techniques to enhance the user experience and network utilization for video delivery. Video optimization is typically “out-of-network” and includes techniques such as transcoding, transrating, time shifting and pacing that are implemented outside of mobile radio access and core networks. However, such techniques cannot easily exploit information about real time cell congestion and radio conditions for the video terminals because it is difficult to obtain such information from the radio access network. We propose a novel “in-network” video optimization technique, named Adaptive Guaranteed Bit Rate (AGBR), for HTTP-based Adaptive Streaming (HAS) video. This optimization technique is implemented at the base station and can thus exploit knowledge of the radio and congestion conditions. With only limited knowledge of the video stream properties or content, AGBR works by adjusting the throughput delivered to the different HAS clients that in turn adjust the video quality they request. The optimization algorithm maximizes aggregate quality across multiple video flows served by the base station without starving data clients, thereby improving the overall quality of experience. We demonstrate through extensive analytical modeling and simulations that AGBR can adapt to changing network conditions to support more video sessions at an acceptable quality than alternative algorithms, while enforcing fairness among all users competing for resources within a sector.

Video streaming, in particular, hypertext transfer protocol based (HTTP) adaptive streaming (HAS) of video, is expected to be a dominant application over mobile networks in the near future. The observation that the base station can alter the video quality requested by a HAS client to its server by controlling the over-the-air throughput from the base station to the client implies that the base station can jointly maximize aggregate video quality of all the HAS flows and throughput of data flows that it serves. We formulate a utility maximization problem that separately takes into account different utility functions for video and data flows and show that the utility maximization can be achieved through an algorithm, we term adaptive guaranteed bit rate (AGBR), wherein target bit rates are calculated for each HAS flow and passed on to an underlying minimum rate proportional fair scheduler that schedules resources across all the flows. This approach has the advantage that it retains the existing scheduling function in the base station with a separate function to compute the target bit rates for the video flows allowing them to only change slowly over time in order to avoid frequent video quality changes. Through analytical modeling and simulations we show that the proposed algorithm can achieve required fairness among the video flows as well as automatically and fairly adapt video quality with increasing congestion thereby preventing data flow throughput starvation.

HTTP adaptive streaming (HAS) is quickly becoming a popular mechanism for delivering on-demand video content over the Internet. The chunked transmission and application-layer adaptation create a very different traffic pattern than traditional progressive video downloads where the entire video is downloaded with a single request. In this paper, we investigate experimentally the interplay between HAS and the network transport control protocol (TCP). We investigate the impact of network delay on achievable throughput and discover that HAS streams cannot fully utilize the available bandwidth due to the start and stop nature of HAS traffic patterns and its interaction with TCP. We investigate TCP pacing as a potential solution to this issue, particularly for packet losses that occur as a result of bursting packets into the network at the start of a transmission. We find that pacing can significantly increase a TCP flow's congestion window but it does not necessarily translate into higher throughput. Instead, we find that packet losses at the end of chunk transmission have a greater impact on throughput.

The consumer interest in mobile multimedia content is on the rise, driven by the higher bandwidth of 3G networks and by the availability of low-cost high-resolution mobile devices. However, providing a good user experience remains a challenge due to bandwidth bottlenecks at peak time, channel quality variations and high battery drain incurred by long data transmission times at lower bandwidths. Consequently high jitter, buffering delays and frequent network outages are ever so common for mobile multimedia services. The emerging mobile broadcast networks (e.g. BCMCS, MediaFLO, DVB-H) are well suited for efficient delivery of highly popular content but lack the on-demand, interactive and retransmission (for reliability) capabilities by virtue of being one-way. In addition, due to business reasons (e.g. MediaFLO) or technical reasons (e.g. BCMCS) service providers prefer to deliver only a limited number of popular channels over these networks. In this paper we propose a mobile content delivery architecture that takes wireless specifics into account to enhance the user experience with multimedia services. Our solution makes efficient use of the available bandwidth, does network and channel quality- aware content delivery on the unicast 3G network while at the same time efficiently and reliably schedules content delivery over the mobile broadcast network. In our solution the delivered content is pre-cached on the storage available on the mobile device. This provides a better user experience, reduces the peak load on the network, and reduces the battery drain on the mobile device. Motivated by the proposed architecture we study the problem of scheduling content over a hybrid unicast and broadcast mobile network and design efficient algorithms and heuristics for the problem.