Search for: All records

Training deep learning (DL) models in the cloud has become a norm. With the emergence of serverless computing and its benefits of true pay-as-you-go pricing and scalability, systems researchers have recently started to provide support for serverless-based training. However, the ability to train DL models on serverless platforms is hindered by the resource limitations of today's serverless infrastructure and DL models' explosive requirement for memory and bandwidth. This paper describes FuncPipe, a novel pipelined training framework specifically designed for serverless platforms that enable fast and low-cost training of DL models. FuncPipe is designed with the key insight that model partitioning can be leveraged to bridge both memory and bandwidth gaps between the capacity of serverless functions and the requirement of DL training. Conceptually simple, we have to answer several design questions, including how to partition the model, configure each serverless function, and exploit each function's uplink/downlink bandwidth. In particular, we tailor a micro-batch scheduling policy for the serverless environment, which serves as the basis for the subsequent optimization. Our Mixed-Integer Quadratic Programming formulation automatically and simultaneously configures serverless resources and partitions models to fit within the resource constraints. Lastly, we improve the bandwidth efficiency of storage-based synchronization with a novelmore »pipelined scatter-reduce algorithm. We implement FuncPipe on two popular cloud serverless platforms and show that it achieves 7%-77% cost savings and 1.3X-2.2X speedup compared to state-of-the-art serverless-based frameworks.« less

Video streaming commonly uses Dynamic Adaptive Streaming over HTTP (DASH) to deliver good Quality of Experience (QoE) to users. Videos used in DASH are predominantly encoded by single-layered video coding such as H.264/AVC. In comparison, multi-layered video coding such as H.264/SVC provides more flexibility for up- grading the quality of buffered video segments and has the potential to further improve QoE. However, there are two challenges for us- ing SVC in DASH: (i) the complexity in designing ABR algorithms; and (ii) the negative impact of SVC’s coding overhead. In this work, we propose a deep reinforcement learning method called Grad for designing ABR algorithms that take advantage of the quality up- grade mechanism of SVC. Additionally, we quantify the impact of coding overhead on the achievable QoE of SVC in DASH, and propose jump-enabled hybrid coding (HYBJ) to mitigate the impact. Through emulation, we demonstrate that Grad-HYBJ, an ABR algo- rithm for HYBJ learned by Grad, outperforms the best performing state-of-the-art ABR algorithm by 17% in QoE.

In an input-queued switch, a crossbar schedule, or a matching between the input ports and the output ports needs to be computed for each switching cycle, or time slot. It is a challenging research problem to design switching algorithms that produce high-quality matchings yet have a very low computational complexity when the switch has a large number of ports. Indeed, there appears to be a fundamental tradeoff between the computational complexity of the switching algorithm and the quality of the computed matchings. Parallel maximal matching algorithms (adapted for switching) appear to be a sweet tradeoff point in this regard. On one hand, they provide the following performance guarantees: Using maxi- mal matchings as crossbar schedules results in at least 50% switch throughput and order-optimal (i.e., independent of the switch size 𝑁 ) average delay bounds for various traffic arrival processes. On the other hand, their computational complexities can be as low as 𝑂 (log2 𝑁 ) per port/processor, which is much lower than those of the algorithms for finding matchings of higher qualities such as maximum weighted matching. In this work, we propose QPS-r, a parallel iterative switching algorithm that has the lowest possible computational complexity: 𝑂(1) per port. Yet,more »the matchings that QPS-r computes have the same quality as maximal matchings in the following sense: Using such matchings as crossbar schedules results in exactly the same aforementioned provable throughput and delay guarantees as using maximal matchings, as we show using Lyapunov stability analysis. Although QPS-r builds upon an existing add-on technique called Queue-Proportional Sampling (QPS), we are the first to discover and prove this nice property of such matchings. We also demon- strate that QPS-3 (running 3 iterations) has comparable empirical throughput and delay performances as iSLIP (running log 𝑁 itera- 2 tions), a refined and optimized representative maximal matching algorithm adapted for switching.« less