Real-time imaging sonar is crucial for underwater monitoring where optical sensing fails, but its use is limited by low uplink bandwidth and severe sonar-specific artifacts (speckle, motion blur, reverberation, acoustic shadows) affecting up to 98% of frames. We present SCOPE, a self-supervised framework that jointly performs compression and artifact correction without clean–noise pairs or synthetic assumptions. SCOPE combines (i) Adaptive Codebook Compression (ACC), which learns frequency-encoded latent representations tailored to imaging sonar, with (ii) Frequency-Aware Multiscale Segmentation (FAMS), which decomposes frames into low-frequency structure and sparse high-frequency dynamics while suppressing rapidly fluctuating artifacts. A hedging training strategy further guides frequency-aware learning using low-pass proxy pairs generated without labels. Evaluated on months of in-situ ARIS sonar data, SCOPE achieves a structural similarity index (SSIM) of 0.77, representing a 40% improvement over prior self-supervised denoising baselines, at bitrates down to ≤ 0.0118 bpp. It reduces uplink bandwidth by more than 80% while improving downstream detection. The system runs in real time, with 3.1 ms encoding on an embedded GPU and 97 ms full multi-layer decoding on the server end. SCOPE has been deployed for months in three Pacific Northwest rivers to support real-time salmon enumeration and environmental monitoring in the wild. Results demonstrate that learning frequency-structured latents enables practical, low-bitrate sonar streaming with preserved signal details under real-world deployment conditions.

Wild salmon are essential to the ecological, economic, and cultural sustainability of the North Pacific Rim. Yet climate variability, habitat loss, and data limitations in remote ecosystems that lack basic infrastructure support pose significant challenges to effective fisheries management. This project explores the integration of multimodal foundation AI and expert-in-the-loop frameworks to enhance wild salmon monitoring and sustainable fisheries management in Indigenous rivers across Pacific Northwest. By leveraging video and sonar-based monitoring, we develop AI-powered tools for automated species identification, counting, and length measurement, reducing manual effort, expediting delivery of results, and improving decision-making accuracy. Expert validation and active learning frameworks ensure ecological relevance while reducing annotation burdens. To address unique technical and societal challenges, we bring together a cross-domain, interdisciplinary team of university researchers, fisheries biologists, Indigenous stewardship practitioners, government agencies, and conservation organizations. Through these collaborations, our research fosters ethical AI co-development, open data sharing, and culturally informed fisheries management.

Large Language Models (LLMs) have become increasingly popular, transforming a wide range of applications across various domains. However, the real-world effectiveness of their query cache systems has not been thoroughly investigated. In this work, we for the first time conducted an analysis on real-world human-to-LLM interaction data, identifying key challenges in existing caching solutions for LLM-based chat services. Our findings reveal that current caching methods fail to leverage semantic connections, leading to inefficient cache performance and extra token costs. To address these issues, we propose SCALM, a new cache architecture that emphasizes semantic analysis and identifies significant cache entries and patterns. We also detail the implementations of the corresponding cache storage and eviction strategies. Our evaluations show that SCALM increases cache hit ratios and reduces operational costs for LLMChat services. Compared with other state-of-the-art solutions in GPTCache, SCALM shows, on average, a relative increase of 63% in cache hit ratio and a relative improvement of 77% in tokens savings.

Sonar radar captures visual representations of underwater objects and structures using sound wave reflections, making it essential for exploration, mapping, and continuous surveillance in wild ecosystems. Real-time analysis of sonar data is crucial for time-sensitive applications, including environmental anomaly detection and in-season fishery management, where rapid decision-making is needed. However, the lack of both relevant datasets andpre-trained DNN models, coupled with resource limitations in wild environments, hinders the effective deployment and continuous operation of live sonar analytics.

Virtualization has attained mainstream status in enterprise IT industry. Despite its widespread adoption, it is known that virtualization also introduces non-trivial overhead when tasks are executed on a virtual machine (VM). In particular, a combined effect from device virtualization overhead and CPU scheduling latency can cause performance degradation when computation intensive tasks and I/O intensive tasks are co-located on a VM. Such an interference also causes extra energy consumption. In this paper, we present Hylics, a novel solution that enables efficient data traverse paths for both I/O and computation intensive workloads. This is achieved with the provision of in-memory file system and network service at the hypervisor level. Several important design issues are pinpointed and addressed during our prototype implementation, including efficient intermediate data sharing, network service offloading, and QoS-aware memory usage management. Based on our real-world deployment on KVM, we show that Hylics can significantly improve computation and I/O performance for hybrid workloads. Moreover, this design also alleviates the existing virtualization overhead and naturally optimizes the overall energy efficiency.

Virtualized datacenters are often designed from scratch with multiple, redundant paths. Yet the majority of the existing congestion control schemes for virtual machines or containers are variants based on single-path TCP design. Lacking the flexibility of leveraging underlying paths, these schemes cannot further improve the utilization of datacenter networks, or mitigate hotspot links. In this paper, we examine the performance of multipath congestion control design on typical hypervisor and container virtualization platforms. We observe that, the involvement of virtual switch, together with the multi-tenancy nature on these platforms, poses new challenges when handling multipath traffic. Through realworld experiments with production-grade applications, we further reveal that, while multipath congestion control increases per-connection throughput and achieves better traffic balancing, it experiences performance degradation when the number of connections is abruptly increased or there exist path-sharing subflows. These issues are due to the enforced QoS policies and interface mapping schemes applied by virtual switch. To this end, we present vMCC, a practical solution which incorporates explicit congestion notification (ECN) support on virtual switches and ECN-aware multipath congestion control algorithms. We show by comprehensive evaluations that vMC-C improves throughput, round trip time, fairness, and energy efficiency for cloud datacenter traffic, and subsequently benefits typical cloud workloads.

Multipath TCP (MPTCP) has recently been suggested as a promising transport protocol to boost the utilization of underlaying datacenter networks, yet it also increases the host CPU power consumption. It remains unclear whether datacenters can indeed benefit from using MPTCP from the perspective of energy efficiency. Through realworld measurement of MPTCP, we show that the energy efficiency of MPTCP is largely related to the flow completion time and the existence of link-sharing subflows. In particular, we find that the link-sharing subflows in MPTCP will significantly elevate the CPUs' power consumption on hosts. To make the matter worse, it will also reduce the transmission efficiency for both throughput-sensitive long flows and latency-sensitive short flows. To address such a problem, we present MPTCP-D, an energy-efficient enhancement of MPTCP in datacenter networks. MPTCP-D incorporates a novel congestion control algorithm that improves energy efficiency by minimizing the flow completion time. It also has a build-in subflow elimination mechanism that precludes link-sharing subflows from increasing the host CPU power consumption. We implement MPTCP-D in the Linux kernel, analyze the parameter selection in the algorithm and study its performance through packet-level simulation and on Amazon EC2. Our results show that, without degrading the performance of the long flow throughput and the short flow completion time, MPTCP-D reduces the long flow energy consumption by up to 72% compared to DCTCP for data transfers, and reduces the short flow power consumption by up to 46% compared to MPTCP with link-sharing subflows.

Nearly 50% of the energy overhead in today's datacenters comes from host-to-host data transfers, which largely depend on the transport layer performance. Multipath TCP (MPTCP) has recently been suggested as a promising transport protocol to improve datacenter network throughput, yet it also increases the host CPU power consumption. It remains unclear whether datacenters can indeed benefit from using MPTCP from the perspective of energy efficiency. By analyzing the performance of MPTCP, we show that (1) despite consuming higher host CPU power than TCP, MPTCP can largely reduce the long flow completion time and thus save the aggregated energy; (2) link-sharing subflows in MPTCP not only has negative impact on both throughput-sensitive long flows and latency-sensitive short flows, but also noticeably increases the host CPU power, especially for short flows. We present MPTCP-D, an energy-efficient variant of multipath TCP for datacenters. MPTCP-D incorporates a novel congestion control algorithm that can provide energy efficiency by minimizing the flow completion time, and an extra subflow elimination mechanism that can preclude link-sharing subflows from increasing the host CPU power. We implement MPTCP-D in the Linux kernel and study its performance by experiments on Amazon EC2. Our results show that, without degrading the performance of the long flow throughput and short flow completion time, MPTCP-D reduces the long flow energy consumption by up to 72% compared to DCTCP for data transfers, and reduces the short flow power consumption by up to 46% compared to MPTCP with linksharing subflows.

Virtualization has become a building block for modern IT industry, and many datacenters are now highly virtualized. It is known that virtualization also introduces non-trivial overhead, which can cause severe self-interference inside a VM when CPU intensive tasks and bandwidth intensive tasks are co-located. Energy efficiency of the server can be affected as well. While such overhead is well-studied in application/protocol specific context, a more comprehensive solution is yet to be explored for general cloud services. In this paper, we present MemNet, a novel protocol-independent solution that enables para-virtualized memory sharing between host and guest VMs. This design successfully decouples I/O and computation operations and lifts the offered interface from the physical devices to high-level network services. Our real-world implementation on KVM indicates that, MemNet can achieve 27% and 70% gain in terms of computing and networking performance, respectively, by resolving the self-interference. It also provides 32% improvement in terms of energy efficiency.

Networking and virtualization are two key building blocks of modern cloud computing. The energy consumption of physical machines has been carefully examined in the past research, including the impact of network traffic. When it comes with virtual machines, the inter-play between energy consumption and network traffic however becomes much more complicated. The traffic are now generated by and exchanged between virtual machines (VMs), which could reside in different physical machines with their respective network interface cards (NICs), or share the same physical machine. When multiple VMs share a physical NIC, their traffic can interfere with each other, causing extra overhead. Yet the VM's allocation can be dynamic and they can even migrated across physical machines, thereby changing the traffic pattern. These factors combined make the network traffic highly diverse and dynamic, so is the corresponding energy consumption. A close examination on the network traffic and energy consumption in virtualized environments is thus of need. In this paper, we present an initial measurement study on the interplay between energy consumption and network traffic in representative virtualization environments. Our study reveals a series of unique energy consumption patterns of the network traffic in this context. We show that state-of-the-art virtualization designs noticeably increase the demand of CPU resources when handling networked transactions, generating excessive interrupt requests with ceaselessly context switching, which in turn increases energy consumption. Even when the physical machine is in an idle state, the VM network transactions will will incur remarkable energy consumption. Furthermore, even with identical number of VMs and amount of traffic on a physical machine, the energy consumptions vary significantly with different VM allocation strategies. Our close examination pinpoints the root cause, and offers new angles to revisit the existing resource usage and energy consumption models, so as to optimize the service provisioning as well as virtual machine placement and migration.