Joshua B. Gahm

Streaming video over HTTP with consistent quality

In conventional HTTP-based adaptive streaming (HAS), a video source is encoded at multiple levels of constant bitrate representations, and a client makes its representation selections according to the measured network bandwidth. While greatly simplifying adaptation to the varying network conditions, this strategy is not the best for optimizing the video quality experienced by end users. Quality fluctuation can be reduced if the natural variability of video content is taken into consideration. In this work, we study the design of a client rate adaptation algorithm to yield consistent video quality. We assume that clients have visibility into incoming video within a finite horizon. We also take advantage of the client-side video buffer, by using it as a breathing room for not only network bandwidth variability, but also video bitrate variability. The challenge, however, lies in how to balance these two variabilities to yield consistent video quality without risking a buffer underrun. We propose an optimization solution that uses an online algorithm to adapt the video bitrate step-by-step, while applying dynamic programming at each step. We incorporate our solution into PANDA -- a practical rate adaptation algorithm designed for HAS deployment at scale.

Fixing multi-client oscillations in HTTP-based adaptive streaming: A control theoretic approach

In recent years, the technology for video delivery over the Internet is shifting towards a new paradigm: HTTP-based adaptive streaming (HAS). An HAS client receives video contents on a segment by segment basis via standard HTTP GET requests. It can dynamically change the rate and quality of the video in the presence of time-varying bandwidth changes. When multiple clients compete over a common bottleneck link, however, they often fail to converge to their respecitive fair share of bandwidth. This leads to constant oscillations in the received video quality. In this paper, we uncover the cause of such oscillations based on observations from large-scale test bed experiments. We then propose a novel client rate adaptation algorithm, which strives to stabilize the playout buffer at a reference level via a proportional-integral controller (PIC). Test bed evaluation results confirm the effectiveness of the proposed PIC scheme and its superior performance over Microsoft Smooth Streaming.

Erasure Coded Storage on a Changing Network: The Untold Story

As faster storage devices become commercially viable alternatives to disk drives, the network is increasingly becoming the bottleneck in achieving good performance in distributed storage systems. This is especially true for erasure coded storage, where the reconstruction of lost data can significantly encumber the system. Thus, a significant amount of research has focused on reducing the amount of data transferred during this repair process. However, in most cases the network is assumed to have a uniform static structure. One reason behind this is that many of the state of the art codes have a fixed repair mechanism or are constrained in the choice of repair strategies, therefore in theory benefit less from being network aware. We propose a general mechanism that explores the space of possible repairs and examine how much different types of erasure codes benefit by being network aware. We show significant gains for three erasure codes using both theoretical modeling and simulation results. We also consider the practical applicability of our proposed mechanism by limiting the search space to repairs that have the potential to be minimal cost and present a case study for RLNC, a class of flexible codes.

Scaling server-based channel-change acceleration to millions of IPTV subscribers

Channel-change times in multicast-based IPTV distribution networks are often slow compared to analog video networks where the channel change is almost instant. One way to overcome these relatively slower channel change is server-based channel-change acceleration based on the rapid acquisition method standardized by the IETF. Concerns about potentially large backbone traffic and bandwidth requirements, however, have prevented operators from deploying such a solution on a large scale. In this paper, we explain techniques to overcome this limit and present data from a real-life deployment. We show how this solution is deployed in a scalable way to almost a million of IPTV set-top boxes while limiting the required backbone traffic and server capacity.

Network-Aware Feasible Repairs for Erasure-Coded Storage

A significant amount of research on using erasure coding for distributed storage has focused on reducing the amount of data that needs to be transferred to replace failed nodes. This continues to be an active topic as the introduction of faster storage devices looks to put an even greater strain on the network. However, with a few notable exceptions, most published work assumes a flat, static network topology between the nodes of the system. We propose a general framework to find the lowest cost feasible repairs in a more realistic, heterogeneous and dynamic network, and examine how the number of repair strategies to consider can be reduced for three distinct erasure codes. We devote a significant part of the paper to determining the set of feasible repairs for random linear network coding (RLNC) and describe a system of efficient checks using techniques from the arsenal of dynamic programming. Our solution involves decomposing the problem into smaller steps, memorizing, and then reusing intermediate results. All computationally intensive operations are performed prior to the failure of a node to ensure that the repair can start with minimal delay, based on up-to-date network information. We show that all three codes benefit from being network aware and find that the extra computations required for RLNC can be reduced to a viable level for a wide range of parameter values.

OpenCL-based erasure coding on heterogeneous architectures

Erasure coding, Reed-Solomon coding in particular, is a key technique to deal with failures in scale-out storage systems. However, due to the algorithmic complexity, the performance overhead of erasure coding can become a significant bottleneck in storage systems attempting to meet service level agreements (SLAs). Previous work has mainly leveraged SIMD (single-instruction multiple-data) instruction extensions in general purpose processors to improve the processing throughput. In this work, we exploit state-of-art heterogeneous architectures, including GPUs, APUs, and FPGAs, to accelerate erasure coding. We leverage the OpenCL framework for our target heterogeneous architectures and propose code optimizations for each target architecture. Given their different hardware characteristics, we highlight the different optimization strategies for each of the target architectures. Using the throughput metric as the ratio of the input file size over the processing latency, we achieve 2.84 GB/s on a 28-core Xeon CPU, 3.90 GB/s on an NVIDIA K40m GPU, 0.56 GB/s on an AMD Carrizo APU, and 1.19 GB/s (5.35 GB/s if only considering the kernel execution latency) on an Altera Stratix V FPGA, when processing a 836.9MB zipped file with a 30×33 encoding matrix. In comparison, the single-thread code using the Intels ISA-L library running on the Xeon CPU has the throughput of 0.13 GB/s.