Bundle adjustment (BA) is a critical technique in various robotic applications such as simultaneous localization and mapping (SLAM), augmented reality (AR), and photogrammetry. BA optimizes parameters such as camera poses and 3D landmarks to align them with observations. With the growing importance of deep learning in perception systems, there is an increasing need to integrate BA with deep learning frameworks for enhanced reliability and performance. However, widely-used C++-based BA libraries, such as GTSAM, g$^2$o, and Ceres Solver, lack native integration with modern deep learning libraries like PyTorch. This limitation affects their flexibility, ease of debugging, and overall implementation efficiency. To address this gap, we introduce an eager-mode BA library seamlessly integrated with PyTorch with high efficiency. Our approach includes a sparsity-aware auto-differentiation design and GPU-accelerated sparse operations designed for 2nd-order optimization. Our eager-mode BA on GPU demonstrates substantial runtime efficiency, achieving an average speedup of 18.5$\times$, 22$\times$, and 23$\times$ across all benchmarks compared to GTSAM, g$^2$o, and Ceres, respectively.

The trade-off of throughput versus latency is important in serving deep neural networks (DNNs) on GPUs. A hardware feature---Multi-Instance GPU (MIG)---provides an additional dimension to this trade-off. This paper studies the GPU sharing problem for serving DNNs with MIG. We propose a system Jormungandr that allocates MIG-enabled GPUs to optimize a user-define utility function, and places DNN serving containers on heterogeneous MIG instances. Jormungandr introduces a utility-first search that yields efficient solutions in practice. Additionally, it presents a robust transition protocol that guarantees a seamless switch between two GPU-cluster configurations, minimizing throughput fluctuations. We implement Jormungandr on Kubernetes. Our experiments show that Jormungandr provides near-optimal solutions, using < 6% more GPUs than the optimal solutions.

Integrating Low Earth Orbit (LEO) satellites with terrestrial network infrastructures to support ubiquitous Internet service coverage has recently received increasing research momentum. One distinct challenge is the frequent topology change caused by the constellation behaviour of LEO satellites. In the context of software defined networking (SDN), the controller function that is originally required to control the conventional data plane fulfilled by terrestrial SDN switches will need to expand its responsibility to cover their counterparts in the space, namely LEO satellites that are used for data forwarding. As such, seamless integration of the fixed control plane on the ground and the mobile data plane fulfilled by constellation LEO satellites will become a distinct challenge. In this paper, we propose the Virtual Data-Plane Addressing (VDPA) Scheme by leveraging IP addresses to represent virtual switches at the fixed space locations which are periodically instantiated by the nested LEO satellites traversing them in a predictable manner. With such a scheme the changing data-plane network topology incurred by LEO satellite constellation can be made completely agnostic to the control plane on the ground, thus enabling a native approach to supporting seamless communication between the two planes. Our testbed-based experiment results prove the technical feasibility of the proposed VDPA-based flow rule manipulation mechanism in terms of data plane performance.

In the inter-domain routing infrastructure, a route leak is defined as a violation of the routing policy agreed between two Autonomous Systems (AS). Route leaks have resulted in large-scale outages on the Internet, taking down several services. Although route leaks seem a simple problem, the solution is complex because: (i) ASes consider -partially- routing policy private, (ii) lack of a formal and standard language to express routing policy and (iii) BGP lacks adequate cryptographic-based security. In this paper, we present an experimental analysis of a distributed ledger-based architecture that provides a solution to route leaks. Specifically, the routing policy is unambiguously expressed using a formal language, that is then stored in a blockchain. This decentralized architecture allows private policies and interfaces seamlessly with the current BGP infrastructure, requiring no changes to routers. We build a prototype to evaluate our proposed architecture using Hyperledger, we analyze its performance using a real-world BGP dataset. Our results show that our architecture scales linearly with relevant metrics. Additionally, we validate the architecture preventing an artificially introduced route leak in a realistic 10 AS topology.

The launch of the StarLink Project has recently stimulated a new wave of research on integrating Low Earth Orbit (LEO) satellite networks with the terrestrial Internet infrastructure. In this context, one distinct technical challenge to be tackled is the frequent topology change caused by the constellation behaviour of LEO satellites. Frequent change of the peering IP connection between the space and terrestrial Autonomous Systems (ASes) inevitably disrupts the Border Gateway Protocol (BGP) routing stability at the network boundaries which can be further propagated into the internal routing infrastructures within ASes. To tackle this problem, we introduce the Geosynchronous Network Grid Addressing (GNGA) scheme by decoupling IP addresses from physical network elements such as a LEO satellite. Specifically, according to the density of LEO satellites on the orbits, the IP addresses are allocated to a number of stationary "grids" in the sky and dynamically bound to the interfaces of the specific satellites moving into the grids along time. Such a scheme allows static peering connection between a terrestrial BGP speaker and a fixed external BGP (e-BGP) peer in the space, and hence is able to circumvent the exposure of routing disruptions to the legacy terrestrial ASes. This work-in-progress specifically addresses a number of fundamental technical issues pertaining to the design of the GNGA scheme.

The Domain Name System (DNS) is one of the most fundamental infrastructures of the Internet. However, due to its design philosophy and implementation architecture, the current DNS still suffers from the centralization problem and the data authenticity problem. In this paper, we analyze these two problems and propose a blockchain-based naming system called blockDNS to solve them simultaneously. In blockDNS, domain names can be applied and transferred freely in a decentralized way. Moreover, a lightweight verification mechanism is also proposed coupled with blockDNS. The verification mechanism allows website clients to verify the authenticity of resolution results with few overheads. Simulation results show that, compared to the Simplified Payment Verification method, blockDNS can cut down the overheads for data authenticity verification from 4.955KBytes to 380Bytes. Moreover, we also present an implementation case of blockDNS, which is compatible with the current naming system and can be deployed incrementally.

Although a series of decoupled or single loop methods have been developed for reliability-based design optimization (RBDO) problems to improve the computational efficiency, it seems hard to extend these strategies to time-variant RBDO due to the complexity of the problems brought by the involvement of time. This paper proposes a new approach for time-variant reliability-based design optimization, expecting to provide an efficient tool for design of some complex structure under time-variant uncertainties. The main idea of the proposed method is the definition of the equivalent most probable point (EMPP). With the EMPP, the original time-variant RBDO problem can be transformed into an equivalent time-invariant RBDO problem formulated by performance measure approach (PMA). Hence, the existing PMA-based time-invariant RBDO methods can be applied to solve the equivalent problem. Therefore, those RBDO methods can be easily extended to time-variant RBDO problems, and hence the computational cost can be effectively reduced. Finally, two numerical examples and an engineering application are used to demonstrate the effectiveness of the proposed method.

In the process of long-term use, the uncertainty of an engineering structure often presents time-variant or dynamic characteristics due to the influence of stochastic loads and material performance degradations. In such a situation, the structural design optimization will involve an important problem of time-variant reliability-based design optimization (TRBDO). Performing TRBDO involves a nested optimization, which will lead to extremely low computational efficiency. In this paper, a single-loop approach (SLA) is proposed to convert the nested optimization in TRBDO into a sequence iterative process composed of the time-variant reliability analysis (TRA), constraint discretization, and design optimization. In each iteration step, the TRA method based on stochastic process discretization is first used to calculate the time-variant reliability of constraints; second, through introducing the concept of the target reliability index of discretized time period and proposing the corresponding algorithm, each time-variant constraint is discretized into a series of time-invariant constraints to formulate a conventional reliability-based design optimization problem. The approach exhibits a good comprehensive performance in terms of efficiency and convergence. The validity and practicality of the SLA are validated by two numerical examples and a design problem for the chassis of a self-balancing vehicle.

Tunneling technology has been widely applied in the network, but the lack of effective congestion exposure in the tunnel, seriously affect the performance of the tunnel technology. This document will focus on the tunnel scenario, to design a tunnel congestion exposure and feedback model, and the tunnel congestion marking scheme and congestion information feedback scheme. First, the document describes Tunneling Protocol, Secondly describes the congestion problems in the tunnel, Thirdly proposes a basic tunnel congestion exposure model, Finally, proposes the detail of tunnel congestion marking and feedback scheme for the model. Modifying the ECN tunnel rules of RFC3168 and RFC6040, congestion marking scheme can be applied to the feedback model. Congestion feedback scheme use GRE (Generic Routing Encapsulation) header format extension to carry the feedback information. Content and format of Congestion feedback, and the transmission of feedback information are also discussed in the document.