INFOCOM 2023 Papers

Maximizing quality of experience (QoE) for interactive video streaming has been a long-standing challenge, as its delay-sensitive nature makes it more vulnerable to bandwidth fluctuations. While reinforcement learning (RL) has demonstrated great potential, existing works are either limited by fixed models or require enormous data/time for online adaptation, which struggle to fit time-varying and diverse network states. Driven by these practical concerns, we perform large-scale measurements on WeChat for Business’s interactive video service to study real-world network fluctuations. Surprisingly, our analysis shows that, compared to time-varying network metrics, network sequences exhibit noticeable short-term continuity, sufficient for few-shot learning requirement. We thus propose Fiammetta, the first meta-RL-based bitrate adaptation algorithm for interactive video streaming. Building on the short-term continuity, Fiammetta accumulates learning experiences through offline meta-training and enables fast online adaptation to changing network states through few gradient updates. Moreover, Fiammetta innovatively incorporates a probing mechanism for real-time monitoring of network states, and proposes an adaptive meta-testing mechanism for seamless adaptation. We implement Fiammetta on a testbed whose end-to-end network follows the real-world WeChat for Business traces. The results show that Fiammetta outperforms prior algorithms significantly, improving video bitrate by 3.6%-16.2% without increasing stalling rate.

We introduce self-adjusting partially ordered lists, a generalization of self-adjusting lists where additionally there may be constraints for the relative order of some nodes in the list. The lists self-adjust to improve performance while serving input sequences exhibiting favorable properties, such as locality of reference, but the constraints must be respected.We design a deterministic adjusting algorithm that operates without any assumptions about the input distribution and without maintaining frequency statistics or timestamps. Despite the more general model, we show that our deterministic algorithm performs closely to optimum (it is 4-competitive). In addition, we design a family of randomized algorithms with improved competitive ratios, handling also a more general rearrangement cost model, scaled by an arbitrary constant d ≥1. Moreover, we observe that different constraints influence the competitiveness of online algorithms, and we shed light on this aspect with a lower bound.We investigate the applicability of our self-adjusting lists in the context of network packet classification. Our evaluations show that our classifier performs similarly to a static list for low-locality traffic, but significantly outperforms Efficuts (by factor 7x), CutSplit (3.6x) and the static list (14x) for high locality and small rulesets.

Federated learning is exposed to model poisoning attacks as compromised clients may submit malicious model updates to pollute the global model. To defend against such attacks, robust aggregation rules are designed for the centralized server to winnow out outlier updates, and to significantly reduce the effectiveness of existing poisoning attacks. In this paper, we develop an advanced model poisoning attack against defensive aggregation rules. In particular, we exploit the catastrophic forgetting phenomenon during the process of continual learning to destroy the memory of the global model. Our proposed framework, called Oblivion, features two special components. The first component prioritizes the weights that have the most influence on the model accuracy for poisoning, which induces a more significant degradation on the global model than equally perturbing all weights. The second component smooths malicious model updates based on the number of selected compromised clients in the current round, adjusting the degree of poisoning to suit the dynamics of each training round. We implement a fully-functional prototype of Oblivion in PLATO, a real-world scalable federated learning framework. Our extensive experiments over three datasets demonstrate that Oblivion can boost the attack performance of model poisoning attacks against unknown defensive aggregation rules.

Flow table is a fundamental and critical component in network data plane. Numerous algorithms and architectures have been devised for efficient flow table construction, lookup, and update. The diversity of flow tables and the difficulty to acquire real data sets make it challenging to give a fair and confident evaluation to a design. In the past, researchers rely on ClassBench and its improvements to synthesize flow tables, which become inadequate for today’s networks. In this paper, we present a new flow table benchmark tool, FlowBench. Based on a novel design methodology, FlowBench can generate large-scale flow tables with arbitrary combination of matching types and fields in a short time, and yet keep accurate characteristics to reveal the real performance of the algorithms under evaluation. The open-source tool facilitates researchers to evaluate both existing and future algorithms with unprecedented flexibility.

Network operators want to enforce fair bandwidth sharing between users without solely relying on congestion control running on end-user devices. However, in edge networks (e.g., 5G), the number of user devices sharing a bottleneck link far exceeds the number of queues supported by today’s switch hardware; even accurately tracking per-user sending rates may become too resource-intensive. Meanwhile, traditional software-based queuing on CPUs struggles to meet the high throughput and low latency demanded by 5G users.We propose Approximate Hierarchical Allocation of Bandwidth (AHAB), a per-user bandwidth limit enforcer that runs fully in the data plane of commodity switches. AHAB tracks each user’s approximate traffic rate and compares it against a bandwidth limit, which is iteratively updated via a real-time feedback loop to achieve max-min fairness across users. Using a novel sketch data structure, AHAB avoids storing per-user state, and therefore scales to thousands of slices and millions of users. Furthermore, AHAB supports network slicing, where each slice has a guaranteed share of the bandwidth that can be scavenged by other slices when under-utilized. Evaluation shows AHAB can achieve fair bandwidth allocation within 3.1ms, 13x faster than prior data-plane hierarchical schedulers.

Secret Sharing (SS) is widely adopted in secure Multi-Party Computation (MPC) with its simplicity and computational efficiency. However, SS-based MPC protocol introduces significant communication overhead due to interactive operations on secret sharings over the network. For instance, training a neural network model with SS-based MPC may incur tens of thousands of communication rounds among parties, making it extremely hard for real-world deployment.To reduce the communication overhead of SS, prior works statically convert interactive operations to equivalent non-interactive operations with extra computation cost. However, we show that such static conversion misses chances for optimization, and further present SOLAR, an SS-based MPC framework that aims to reduce the communication overhead through dynamic communication-computation conversion. At its heart, SOLAR converts interactive operations that involve communication among parties to equivalent non-interactive operations within each party with extra computations and introduces a speculative strategy to perform opportunistic conversion when CPU is idle for network transmission. We have implemented and evaluated SOLAR on several popular MPC applications, and achieved 1.6-8.1 times speedup in multi-thread setting compared to the basic SS and 1.2-8.6 times speedup over static conversion.

Machine learning (ML) tasks are one of the major workloads in today's edge computing networks. Existing edge-cloud schedulers allocate the requested amounts of resources to each task, falling short of best utilizing the limited edge resources flexibly for ML task performance optimization. This paper proposes TapFinger, a distributed scheduler that minimizes the total completion time of ML tasks in a multi-cluster edge network, through co-optimizing task placement and fine-grained multi-resource allocation. To learn the tasks' uncertain resource sensitivity and enable distributed online scheduling, we adopt multi-agent reinforcement learning (MARL), and propose several techniques to make it efficient for our ML-task resource allocation. First, TapFinger uses a heterogeneous graph attention network as the MARL backbone to abstract inter-related state features into more learnable environmental patterns. Second, the actor network is augmented through a tailored task selection phase, which decomposes the actions and encodes the optimization constraints. Third, to mitigate decision conflicts among agents, we novelly combine Bayes' theorem and masking schemes to facilitate our MARL model training. Extensive experiments using synthetic and test-bed ML task traces show that TapFinger can achieve up to 28.6% reduction in the average task completion time and improve resource efficiency as compared to state-of-the- art resource schedulers.

In many applications such as statistical machine learning, reinforcement learning, and optimization for large data centers, the increasing data size and model complexity have made it impractical to run optimizations over a single machine. Therefore, solving the distributed optimization problem has become an important task. In this work, we consider a distributed processing network $G = \left( {\mathcal{V},\mathcal{E}} \right)$ with n nodes, where each node i can only evaluate the values of a local function (i.e., has zeroth-order information) and can only communicate with its neighbors. The objective is to reach consensus on the global optimizer of ${\max _{x \in \mathcal{X}}}\frac{1}{n}\sum\nolimits_{i = 1}^n {{f_i}(x)} $. Previous methods either assume first-order gradient information which is not suitable for many model-free learning scenarios, or consider the zeroth-order information but assume convexity of the objective functions and can only guarantee convergence to a stationary point for nonconvex objectives. To address these limitations, we drop both the known gradient assumption and convexity assumption. Instead, we propose a distributed Bayesian framework for the problem with only zeroth-order information and general nonconvex objective functions in a Matérn Reproducing Kernel Hilbert Space (RKHS). Under this framework, we propose an algorithm and show that with high probability it reaches consensus on all nodes and has a sublinear regret with regard to the global optimal. The results are validated under numerical studies.

Exploiting millimeter-wave (mm-wave) for high-capacity multi-user networks is predicated on jointly performing beam management for seamless connectivity and efficient resource sharing among all users. Beam management in 5G-NR actively monitors candidate beam pair links (BPLs) on the serving cell to simply select the user’s best beam, but neglects the multi-user resource sharing problem, potentially leading to severe throughput degradation on overloaded cells. We propose flexRLM, a coordinator-based flexible radio link monitoring (RLM) framework for multi-user downlink mm-wave networks. flexRLM enables flexible configuration of monitored BPLs on the serving and other candidate cells and beam selection jointly considering link quality and resource sharing. flexRLM is fully 5G-NR-compliant and uses the LTE coordinator in non-standalone mode to continuously update the monitored BPLs via measurement reports from periodic downlink control synchronization signals. We implement flexRLM in ns-3 and present full-stack simulations to demonstrate the superior performance of flexRLM over default 5G-NR RLM in multi-user networks. Our results show that flexRLM’s continuous updating of monitored BPLs improves both link quality and stability. By monitoring BPLs on candidate cells other than the serving one, flexRLM also significantly decreases handover decision delays. Importantly, flexRLM’s low-complexity coordinated load-balancing achieves a per-user throughput close to the single-user baseline.