Papers

Workflow scheduling in user device-centric edge-cloud computing has emerged as a challenging problem due to the need to jointly optimize execution performance, energy consumption, and monetary cost. Modern workflows consist of heterogeneous tasks, where compute-intensive components are well-suited for Infrastructure-as-a-Service (IaaS) resources, while fine-grained, event-driven tasks are more efficiently executed using the Function-as-a-Service (FaaS) model. We formulate this scheduling problem for application workflows over a hybrid User Device (UD)-IaaS-FaaS computing continuum as a multi-objective Mixed-Integer Linear Program (MILP). The formulation models the workflow execution mechanism by incorporating cumulative capacity constraints for IaaS resources and elastic resource provisioning for FaaS. To overcome the intractability of MILP, we propose a learning-based framework that combines graph neural networks, surrogate performance modeling, and a multi-sample REINFORCE approach. The proposed approach is workflow-aware, leveraging task pairing to improve data locality and structured task placement decisions. Agent training is organized into three stages: small-scale learning of UD-IaaS trade-offs, intermediate locality aware co-location, and large-scale handling of parallelism and aggregation bottlenecks in workflows. The proposed framework is validated through real-world experiments using the Pegasus Montage workflow on a custom UD-IaaS-FaaS testbed comprising a user device, multiple IaaS nodes, and serverless FaaS instances. Extensive experimental results demonstrate the improved trade-offs across makespan, energy consumption, and cost. Compared with the genetic algorithm and greedy scheduling approaches, the proposed approach reduces makespan by 10-17.5%, energy consumption by 3-9%, and cost by 4-10%, while remaining within 3.5-5.6% of the MILP optimum.

This paper proposes a fully autonomous execution model for multi-GPU applications that completely excludes the involvement of the CPU beyond the initial kernel launch. In a typical multi-GPU application, the host serves as the orchestrator of execution by directly launching kernels, issuing communication calls, and acting as a synchronizer for devices. We argue that this orchestration, or control flow path, causes undue overhead and can be delegated entirely to devices to improve performance in applications that require communication among peers. For the proposed CPU-free execution model, we leverage existing techniques such as persistent kernels, thread block specialization, device-side barriers, and device-initiated communication routines to write fully autonomous multi-GPU code and achieve significantly reduced communication overheads. We demonstrate our proposed model on two broadly used iterative solvers, 2D/3D Jacobi stencil and Conjugate Gradient(CG). Compared to the CPU-controlled baselines, the CPU-free model can improve 3D stencil communication latency by 58.8% and provide a 1.63x speedup for CG on 8 NVIDIA A100 GPUs. The project code is available at https://github.com/ParCoreLab/CPU-Free-model.

Modern cloud platforms generate high-dimensional, noisy telemetry from thousands of microservices. Unlike classical anomaly detection benchmarks, cloud anomalies are short-lived, irregular, and largely non-periodic, making timely detection challenging. We present an empirical benchmark of deep learning models for anomaly detection using a real-world IBM Cloud Console dataset. We evaluate GRU, Transformer, TimesNet, and Autoformer architectures using the Numenta Anomaly Benchmark (NAB), a timing-aware metric that accounts for early detection and false alarms. TimesNet achieves the highest NAB scores due to strong early-detection sensitivity. Transformer models also detect anomalies early but incur higher false-positive rates. An improved GRU autoencoder, incorporating robust scoring and persistence rules, achieves competitive performance with a more stable false-alarm profile. Overall, the modest performance differences suggest that real cloud anomalies are dominated by short-term deviations rather than persistent periodic structure, emphasizing the importance of timing-aware evaluation and robust post-processing.

Hybrid attention-SSM serving combines growing attention KV with fixed-size recurrent and convolution state whose allocation units, reuse boundaries, and lifetimes differ. We implement four memory organizations in a modified vLLM runtime: a coarse shared-page baseline (S0), checkpoint-internal recurrent paging (S1), paging with convolution-slot packing (S2), and native dual pools (D0). Nine evaluation questions cover storage costs and serving behavior. The cumulative S0-S1-S2 refinement improves completed-output throughput and tail completion time at the main operating point. D0-S2 performance varies with memory budget, offered load, and admission headroom. Shared-prefix workloads yield larger stepwise gains; gains also persist in a no-sharing control with equal recorded input-work totals. State traces identify recurrent-checkpoint availability as the limit on joint reuse in the tested S2 runs. Extending producer-side checkpoint lifetime increases reusable depth at a fixed page budget. A fresh-seed SHARE/ISOLATE comparison estimates a positive retention interaction, with a primary confidence interval that includes zero. Retained-state continuation correctness remains unresolved. These findings motivate evaluating physical state representation together with request admission and checkpoint lifetime when designing memory management for concurrent hybrid-model serving.

Temporal graph neural networks (TGNNs) achieve superior performance in many dynamic graph learning tasks. These models use node state vectors (i.e., memories) to capture and record graph changes (i.e., events), and then exploit these memories to assist prediction. However, this memory-centric design makes multi-GPU training challenging because predictions for later events often depend on memory updates from earlier events. To maintain model quality, existing multi-GPU TGNN training schemes strictly preserve all event dependencies, which limits parallelism and causes expensive cross-GPU memory synchronization. In this work, we observe that many of these dependencies can be handled more flexibly without harming training quality. Specifically, (a) dependencies in TGNN training have high locality in both time and graph topology, (b) the short-range dependencies missed within a large batch can often be recovered cheaply, and (c) many memory updates are not immediately needed. Based on these observations, we propose DepBloom to accelerate multi-GPU TGNN training with three techniques: (1) temporally bounded spatial partitioning to reduce unnecessary cross-GPU synchronization and improve parallelism; (2) intra-batch memory refinement to recover missing short-range temporal dependencies inside enlarged batches; and (3) critical-path optimization to prioritize urgent memory updates while deferring non-critical work during multi-GPU pipeline execution. Experimental results across representative models and diverse datasets show that DepBloom achieves an average speedup of 2.84$\times$ and up to 5.16$\times$ over state-of-the-art methods, while maintaining comparable or better model accuracy.

Large-scale distributed training increasingly makes collective communication bandwidth-bound. In modern GPU clusters, communication libraries such as NCCL implement collectives including AllReduce over heterogeneous interconnects such as NVLink, InfiniBand, and Ethernet, making efficient use of available network bandwidth a central systems challenge. ForestColl showed that collective communication on such fabrics can be formulated using fractional tree packing; its scalable polynomial-time construction is developed for AllGather, with AllReduce obtained by composing AllGather and ReduceScatter. We present DualColl, which targets bandwidth-bound AllReduce directly. Under the tree-based AllReduce model, DualColl applies a Garg–Könemann-style multiplicative-weights algorithm and, for any $0<\epsilon<1$, computes a feasible schedule whose throughput is at least $(1-\epsilon)^2$ of the optimal AllReduce throughput. Its key component is an exact minimum-cost collective-tree oracle that projects physical-link dual prices onto a logical graph of participating ranks, reducing each iteration to shortest-path computations followed by a minimum spanning tree. We further extend the framework to local equal-next-hop packet-sprayed routing and to networks without explicit per-tree bandwidth control. For the latter, we introduce a tree-coupled max-min bandwidth-sharing model and show that tree duplication approximates an explicit fractional allocation with a provable accuracy-complexity tradeoff. Finally, we extend the fractional-packing framework to additional collective operations and multiple concurrent collective groups.

Can quantum communication determine the size of an anonymous network when no size bound is known? Classical impossibility proofs rely on the covering lemma: an execution can be reproduced on a larger network with the same local structure. Quantum interference invalidates this argument - a measurement record possible on a network can become impossible on a cover. We show that the impossibility of exact counting nevertheless survives. Our main result is a quantum analogue of the covering lemma that preserves local possibilities rather than complete executions. Within a cyclic covering family, every local measurement outcome possible at a node remains possible at a corresponding node in every sufficiently large network. This holds even after information has travelled around the smaller network, beyond the reach of a simple locality argument. We extend the principle to the universal covering tree and finite covers with sufficiently long cycles. The proof combines tensor-network methods with graph coverings: a common representation across network sizes turns normalization into a constraint on which local outcomes can occur. For uniform synchronous protocols with independently prepared nodes, this support-lifting principle rules out zero-error counting and even the production of a guaranteed finite size bound on anonymous rings and the other cyclic covering families considered here. It also rules out exact leader election with termination in finite time with probability one on these families. Measurements, classical control, and local memory growing with time are allowed; no predetermined running-time bound is assumed. Quantum interference thus defeats execution lifting without enabling infallible certification of network size.

We study constant-round LOCAL approximation for additive local vertex-subset minimization problems. Given a $k$-uniform $\alpha$-approximation on a hereditary source class closed under disjoint union, Bonamy, Das, Gavoille, Picavet, Suomela, and Wesolek obtained ratio $\alpha(d+1)+1$ on locally nice target classes of asymptotic dimension $d$. We prove two transfers with ratios $\alpha+1+\eta$ and $\alpha+\eta$ for every $\eta>0$. These bounds require the stronger local niceness assumption at a larger scale that depends on $\eta$. The first algorithm keeps their repair construction and increases the radius used to detect errors. We average over partitions of the graph to bound the retained cost without the factor $d+1$. These partitions are used only in the analysis. The second algorithm tests several deletion widths around the errors. In each component of the repair zone, it chooses a smallest union of retained choices and an exact repair. Averaging over the widths bounds the cost counted twice and yields ratio $\alpha+\eta$. Both transfers are deterministic and computable, use arbitrary distinct numerical identifiers, and need no knowledge of the graph size. Their outputs also satisfy local uniformity guarantees. For 2-domination on graphs of bounded Euler genus, the same planar source gives ratio $19$ through the earlier transfer and $6+\eta$ through ours. For minimum dominating set, our transfer gives ratio $7+\varepsilon$ on graphs formed by attaching a clique of arbitrary size at one vertex of a planar graph, using the source algorithm of Amiri and Siebertz. This class satisfies our target assumptions and allows nonempty error sets. Its Euler genus is unbounded, so their guarantee for graphs of bounded Euler genus does not cover the whole class.

Design-space exploration of LLM inference clusters requires predicting how hardware and serving policies interact with hierarchical key-value (KV) caching. Isolated operation profiles alone do not determine cluster latency: reuse changes computational demand, while KV location, capacity, and contention govern execution order. We present a compositional discrete-event simulator based on separating local service costs from shared resource and KV availability. Compute and transfer models calibrated on one-prefill/one-decode (1P-1D) deployments are reused across compatible workers, enabling policy and topology comparisons without per-configuration latency fitting. Validation against SGLang covers Qwen3-8B and Qwen3-32B on 1P-1D and 3P-1D clusters. Mean request end-to-end latency error is 1.34% for question answering and 6.36% for captured coding sessions, measured as mean absolute percentage error. CPU simulation reduces replay turnaround by 15.33 times on average; real replay includes prescribed arrival gaps. Measured session-affinity and simulated retention studies demonstrate how the framework guides policy selection by predicting the joint effects on computation, latency, and KV occupancy.

Production vector search must continuously ingest, delete, and re-tenant vectors while serving low-latency, high-recall, metadata-filtered approximate nearest-neighbor (ANN) queries for many tenants with per-tenant service-level objectives. Existing graph-based ANN systems address these requirements largely in isolation, leaving recall during concurrent re-optimization without an analytical bound. We develop a graph-based index that combines streaming updates, filtered search, and multi-tenancy, with one theoretical result and two system mechanisms. Our theory decomposes per-tenant recall loss into staleness from updates not yet integrated into the graph and filtered navigability determined by the local density of points satisfying a tenant's predicate. The staleness component exhibits anti-amortization on a merge-based graph, integrating a larger batch of inserts increases recall degradation super-linearly, so larger merges do not reduce total staleness loss. Holding the insert horizon fixed while increasing merge size 20x, we observe a monotonic decrease in terminal recall@10 from 99.44% to 99.01%. For filtered search, the theory gives a parameter-free frontier-mass relationship in local predicate density and a query-averaged per-tenant recall bound. We instantiate the bound with a filter and tenant-aware staging cache and a select-not-resize scheduler with per-tenant staleness ceilings. On a FreshDiskANN substrate, the ceiling eliminates tight-tier SLA violations under the evaluated churn workload, while merge reordering changes terminal recall by only 0.02 percentage points. The results show how recall bounds can guide background maintenance without relying on merge resizing as a recall control mechanism.

Beeping Networks (BN) have become an established model of wireless communication and computation. They assume, however, that the communication interference caused by beeping nodes is binary (OR channel), while both artificial and nature-driven networks are based on varied signal strength and receiving thresholds. This work naturally extends the classic beeping networks to weighted beeping networks (WBN), by modeling signal strengths by weights on graph edges and receiving a beep if the accumulative signal strengths of transmitting neighbors are above receiving threshold. Under this generalized model, we study the fundamental Local Broadcast (LB) problem. Since edges with low weights may not even accumulate to a received beep, we only require that messages are delivered along sufficiently heavy edges, with weights above some given threshold $\sigma$. We design an LB algorithm that completes its task in $O(I_\sigma \log n \cdot (\mathcal{M}+\log n))$ beeping rounds, whp, where $n$ is the number of nodes, $I_\sigma$ is the maximum local interference on heavy edges, and $\mathcal{M}$ is the maximum length of messages sent locally. This result is nearly optimal, as, with probability $1$, $\Omega(I_\sigma)$ beeping rounds are necessary to solve LB. Finally, we show that any algorithm designed for the classic beeping networks and terminating in polynomial number of beeping rounds, can be emulated on any WBN with a logarithmic overhead, that is, it returns the same output as working on the classic beeping network isomorphic to the sub-graph of heavy edges.

Pipeline parallelism has become a critical technique for training large language models. To address the throughput limitation of existing sequence-level methods that schedule subsequences unidirectionally, we propose bidirectional sequence-level interleaved pipeline parallelism (BSIPipe), which co-schedules two opposing V-shaped interleaved pipelines at the subsequence level. To bound the activation accumulation from subsequence scaling, BSIPipe introduces a basic-unit concatenation mechanism that keeps activation memory invariant to the subsequence count. To eliminate communication deadlock risk from bidirectional scheduling, BSIPipe employs a communication optimization strategy that combines topology-aware device mapping with communication-computation overlap. In addition, BSIPipe composes with tensor parallelism to overcome the memory limitation of single-dimensional parallelism, neutralizing its bidirectional memory overhead while maintaining throughput. Experiments on GPT and LLaMA3 demonstrate that BSIPipe achieves average speedups of 1.11$\times$ to 1.13$\times$ in long-sequence training over SOTA sequence-level baselines.

5G edge network slicing must reconcile two coupled requirements: understanding high-level operator intent and reallocating scarce computing and bandwidth resources under dynamic contention. Centralized optimization can face computational and responsiveness challenges under rapidly changing edge contention, and Multi-Agent System (MAS)-based auctions can support resource arbitration but still depend on operator-defined economic parameters and priorities. Embedding a Large Language Model (LLM) directly in a recurring control loop, by contrast, may introduce non-negligible inference latency and non-deterministic behavior. This paper proposes a two-timescale architecture that addresses this gap: an LLM-based Strategist translates natural-language network intents into economic policy parameters (budget, income, and bidding-aggressiveness) at the macro scale, while a deterministic, Vickrey-inspired second-price mechanism run by SPADE-based slice agents resolves recurring resource contention at the operational scale, with the resulting allocations enforced by patching Kubernetes CPU limits and Calico CNI bandwidth annotations directly on User Plane Function pods. We evaluate the architecture end-to-end on a real Kubernetes/Open5GS edge cluster across two complementary scenarios, each repeated 30 times: a Smart-City scenario, in which natural-language public-safety intents propagate into measurable changes in slice bids and bandwidth allocations within a few auction rounds, and a Hierarchical-slicing scenario, in which the mechanism provides priority-dependent resource and QoE differentiation, adapts to abrupt demand changes, and preserves configured minimum guarantees while maintaining near-complete allocation of the shared bandwidth pool.

Industrial IoT sensors produce continuous, high-volume data streams that predictive maintenance needs to analyse in real time. Cloud-centric architectures send all raw telemetry to central servers, which costs latency, bandwidth and resilience. This paper presents a hybrid cloud-edge architecture for manufacturing predictive maintenance built on AWS IoT Core and AWS IoT Greengrass v2, with local machine learning inference at the edge and selective cloud synchronization. We deploy two configurations on real AWS infrastructure and compare them: a cloud-only baseline (Configuration A) and the proposed hybrid edge architecture (Configuration B). We measure end-to-end inference latency, bandwidth consumption and inference accuracy. Configuration B cuts cloud-bound bandwidth by 86.8% (63.6 KB to 7.2 KB per 200-reading window) and decision latency by 256x (23.1 ms cloud round trip to 0.09 ms local inference). At the edge, with no cloud involvement, it keeps an anomaly-class F1 score of 0.84 and 91% recall. The evaluation also instruments system uptime and Stream Manager queuing under cloud disconnection; the quantitative disconnection experiment is left as planned work. Our review of the literature points to four gaps that this work addresses: the lack of empirically validated Greengrass v2 deployments for predictive maintenance, unquantified bandwidth savings from selective edge synchronization, unmeasured resilience under cloud disconnection, and the open MLOps feedback loop at the edge.

Enterprise AI agents can execute altered versions of approved actions, concurrently reuse the same permission, or retry operations whose responses were lost. These failures arise when approval and resource execution lack a shared contract. KedgeFlow specifies language-independent contracts for trusted controllers to authorize state-changing operations and privileged gateways to enforce that authority independently of agent reasoning. Inspired by OpenFlow, it connects policy checks supported by evidence, signed permission for an exact action, a shared allowance that can be consumed only once, target-side conditions for applying the change, and retained outcome records. Under trusted issuance of exact-action authority, two conditional properties establish at most one successful claim per allowance and link every governed dispatch to a matching grant, claim, and subsequent dispatch authorization, even when the agent is compromised. A reference implementation exposes this execution path to agent systems through the Model Context Protocol (MCP). In the single-resource evaluation, 6,560 concurrent ledger claim attempts yielded exactly one successful claim for each of 82 allowances. Boundary tests rejected unauthorized and altered requests before consumption. A 32-call gateway race produced one committed mock effect, and a retry after a lost response produced no additional dispatch. Deliberately permissive policy nevertheless authorized an undesirable effect, exposing the limits of execution authorization. Evaluation uses three members of the etcd coordinated key-value store on one host, synthetic authentication and authorization inputs, and a mock target that rejects stale requests. Distributed reservations, enforcement of every production access route, native crash recovery, exactly-once effects at real targets, and interoperability between independent implementations remain untested.

Agent applications are evolving into workflows composed of large language model (LLM) calls and tool invocations, termed modules. Existing distributed agent systems typically adopt a disaggregated architecture, deploying different modules on separate devices. While this enables pipeline parallelism, imbalanced module latencies introduce severe pipeline bubbles that limit throughput. A natural alternative is an aggregated architecture that co-deploys all modules across all devices, eliminating bubbles by keeping every device continuously active. However, this introduces request blocking: short requests are forced to progress in lockstep with longer ones in the same batch, and newly arriving requests must wait until the entire ongoing batch completes the full workflow. To address this, we propose AgentWeave, an efficient distributed serving system for agent applications built on the aggregated architecture. Its core mechanism, flow decomposition, decomposes long agent requests into fine-grained execution units while preserving agent semantics, allowing short and newly arriving requests to proceed without being stalled by long-running ones. Complementing this, a flow-adaptive scheduler dynamically balances the throughput gains of decomposition against its scheduling overhead. Experiments across four representative agent applications demonstrate that AgentWeave improves throughput by 1.81x, reduces average latency by 2.45x, and reduces P90 latency by 2.04x over state-of-the-art baselines.

Distributed training of sparsely activated mixture-of-experts (MoE) models is often bottlenecked by workload imbalance and all-to-all (A2A) communication induced by dynamic token routing. Existing token deduplication, expert placement, and replication techniques alleviate these bottlenecks but do not jointly capture their distinct load semantics. Under hierarchical token deduplication, communication counts each token once per destination group at each hierarchy level, whereas expert computation processes every token-expert assignment. With replication, both costs depend jointly on physical layout and token-to-copy mapping, making separate optimization suboptimal. We present PlaceMoE to address this coupling. First, we formulate the joint optimization of expert layout and source-aware token-to-copy mapping for E experts distributed across P devices, accounting for both hierarchical communication traffic and expert-compute load. Second, we develop a solver that combines co-selection-aware expert grouping, bounded dynamic programming for replica allocation, and alternating refinement of layout and mapping. Third, PlaceMoE overlaps replica-gradient synchronization with attention backward. We evaluate PlaceMoE on a 32-GPU cluster and a 64-NPU cluster using Qwen3-VL and DeepSeek-V3. Experimental results show that PlaceMoE achieves up to a 6.94x speedup in A2A communication and 1.74-2.33x speedups in end-to-end training over the state-of-the-art training framework VeOmni.

Energy-aware orchestration across geographically distributed infrastructures is increasingly important as data centers absorb growing Artificial Intelligence (AI) workloads. However, existing approaches often exclude the energy cost of automated decision-making from system-level sustainability assessments. This paper presents an energy-aware agentic orchestration framework that treats Large Language Model (LLM) reasoning as a measurable component of the control loop. We compare single-shot tool calling, ReAct, and plan-and-execute strategies on a multi-site Kubernetes testbed spanning two European data centers and instrumented at layers ranging from power distribution units to per-pod energy attribution. Using ten natural-language orchestration intents with varying decision complexity, we show that reasoning strategy and intent complexity strongly influence graphics processing unit (GPU) energy consumption during orchestration decisions. ReAct and plan-and-execute generated executable tool calls across all evaluated intents but required multiple LLM inferences, whereas the single-shot configuration was less costly but non-actionable. Prefix caching reduced mean ReAct GPU energy consumption by 29.4%. For the evaluated lightweight workload, the infrastructure-energy changes observed within the one-second measurement window were insufficient to offset the measured GPU decision energy. These findings demonstrate the importance of accounting for reasoning overhead when assessing the sustainability of AI-native orchestration and motivate energy-budgeted selection of agent strategies.

Grassroots platforms are designed to be owned, operated and governed by their participants, and thus are constituted from person-to-person smartphone apps that are run by the participants. A grassroots app is specified using a grassroots social contract, which is compiled through an abstraction cascade into a volition-guarded Grassroots Logic Program (vGLP), a multiagent concurrent logic programming language. A vGLP program in turn specifies a smartphone UI via which people express their volitions and thus control the app. Prudence dictates large-scale evaluation and testing of grassroots platforms prior to deployment, but using the many people needed to do that is a contradiction in terms. Hence another solution is needed. Here, we present stochastic GLP (sGLP), designed to allow grassroots platforms to be simulated at scale from the vGLP programs that deploy them, with stochastic agents in place of people. sGLP is GLP with a rate extension for specifying stochastic behaviour. A stochastic agent is an sGLP program replacing both the vGLP UI and the person behind it. It is typed by the dual of the interactive type of the UI and specifies a stochastic profile of a person's behaviour. We demonstrate sGLP on the grassroots social graph, with a population of men and women drawn from four stochastic profiles — homophile or indifferent, wary or sociable — and on grassroots currencies, with four profiles — spenders or savers, cautious or generous — and report what the mix of profiles does to the social graph that forms and to the currencies that circulate. We prove that every run of a vGLP program in which sGLP agents replace the UI and the people behind it is a vGLP run. We implemented sGLP in GLP by Gillespie's method, and ran reproducible simulations with it.

Agentic applications generate multi-turn sequences of LLM requests, with each new turn extending the context accumulated from prior interactions and tool calls. Prefix caching reuses the KV cache of earlier turns and avoids redundant prefill, but routing every turn to its cache-affine instance can create severe local queues while capacity remains idle elsewhere. Remote KV cache access relaxes this tension, yet migrating execution adds transfer work and is harmful when the cluster is already congested. Thus, deciding where to execute a request requires reasoning jointly about cache locality, queueing, and the current operating point of the cluster. We present Vesta, a queueing-aware scheduler for distributed agentic LLM serving. Vesta assigns each session a stable holder for its authoritative KV cache while independently selecting an executor for every request. Grounded in classical queueing theory, its objective accounts for backlog, reusable prefixes, KV cache transfer, and the queueing externality of additional service demand; it migrates execution to recover idle capacity when beneficial and favors placements that add the least prefill and transfer work as congestion rises. Vesta also migrates reusable prefixes across length buckets to avoid cold starts. On real-world agentic workloads, Vesta improves throughput by 14.0% and reduces mean TTFT by 57.8% over the state-of-the-art scheduler in a PD-colocated deployment. In PD-disaggregated deployments, it reduces mean TTFT by up to 18.1% and P99 TTFT by up to 29.0% at matched throughput.

Distributed quantum computing distributes workloads that exceed a single processor's capacity across resource-constrained components. Circuit cutting enables this decomposition, but its effectiveness depends on partition choice. We present SACE, a Simulation-Aware Circuit Partitioning framework with Entanglement-Capability-Aware Screening. EP-Cost screens partitions using a parameter-dependent entanglement-capability proxy and native implementation cost. Under reconstruction constraints, finite-bond refinement selects the candidate that minimizes the worst-subcircuit discrepancy between normalized MPS simulations at bond dimensions $\chi$ and $2\chi$. Across 14 exhaustively enumerated small-scale candidate spaces, EP-Cost-guided genetic algorithm selections attain co-optimal native quasiprobability decomposition (QPD) overhead, with a median within-circuit Spearman correlation of 0.9965 between EP-Cost and QPD overhead. At matched cut counts, EP-Cost reduces QPD overhead by $25.20\times$ and $16.89\times$ for 8- and 9-qubit QFT circuits, respectively, and improves sampled reconstruction. On 12-qubit approximate QFT, deterministic reconstruction reduces Pauli-observable mean absolute error in all six low-bond conditions and total-variation distance in five. With five cuts and 7,776 QPD terms, SACE is best on all four reconstruction metrics in four of six conditions. At 24 qubits, SACE improves exact-local QFT accuracy in all 30 instance-bond-dimension conditions against each EP-Cost-only, Kernighan-Lin, and spectral baseline. The 50-qubit QFT comparisons remain favorable as reference convergence improves. Weighted QAOA circuits provide complementary, more bond-sensitive evidence. Refinement benefits persist across three search methods; larger candidate panels improve the optimized proxy without guaranteeing lower exact-local error.

Large Language Model (LLM) inference is driving a growing share of datacenter carbon emissions. Although carbon-aware scheduling can reduce these emissions in geo-distributed datacenters by exploiting the spatiotemporal variations in grid carbon intensity, the latency-sensitive nature of LLM serving constrains this design space. While prior studies investigate the carbon savings achievable through spatial scheduling, they oversimplify the dynamics of modern inference engines, such as queueing and continuous batching, which strongly shape inference latency. We address this issue by designing and evaluating a latency-centric, geo-distributed inference simulator, calibrated with hardware-grounded latency and power measurements, that reproduces these dynamics and enables the study of carbon savings under strict service-level objectives (SLOs). We then formulate a series of SLO- and carbon-aware schedulers of increasing sophistication, from per-request heuristics to batch-level optimization: (i) a greedy heuristic, (ii) an integer linear programming (ILP)-based scheduler jointly assigning batches of requests, (iii) an iterative ILP-based solution projecting its own assignments back into the load estimates to capture load coupling, and (iv) a facility mixed-integer linear programming (MILP) scheduler pricing the energy of batched requests collectively rather than in isolation. Across three LLMs under MLPerf SLOs and varying system loads, the MILP-based option offers the best carbon-SLO balance, reducing carbon emissions by up to 60% relative to the round-robin policy, with an average SLO violation of 3%. Our characterization further shows that savings grow with the spread of carbon intensity across datacenters and shrink as system load increases, while SLO attainment depends on how frequently schedulers receive datacenters' load updates.

Physical systems and distributed AI workflows can retain past information, receive delayed feedback and combine conflicting signals. These features pose a common modeling problem: determining what information about past interactions a state must retain to account for future responses. Active Neural Matter (ANM) addresses this problem through a finite-window response framework. Forward analysis uses each system's dynamics to predict state and output responses to perturbations, separating propagation from output sensitivity. Inverse analysis tests whether a candidate state retains enough information for these responses. With the system and future inputs or interaction rules fixed, distinguishable responses from histories assigned the same state identify distinctions the representation must preserve. Their distributional separation gives a lower bound on the worst-case error of approximating behavior from that state. Uncertainty bounds distinguish rejection, support within a fixed tolerance on tested comparisons, and unresolved cases. Delayed-consensus simulations check analytical predictions of response amplitude and phase over a narrow range of delays; flocking illustrates dependence on the chosen observable. An exact delayed-return construction establishes a sufficient state for a specified program. The AI studies identify distinctions in interaction history and evidence order that a representation must preserve to reproduce observed response distributions. Tests on new histories in a controlled feedback task provide bounded support for a state specified by the task instructions; the instructed rule predicts every observed action. ANM connects response prediction to measurable information requirements, separating behavioral fidelity from task correctness and initial-state adequacy from a closed evolution law for that state.

To operate distributed applications with high availability, operators must rapidly diagnose and mitigate any incident. LLM-based agents can help, but unlike experienced human operators they do not improve as they handle more incidents. Each investigation starts from scratch, re-deriving likely causes and re-testing hypotheses even when a similar fault was resolved in the past. We present a self-defining operator (SDO) that improves with time by updating its operational behavior and memory. When SDO resolves an incident, it distills that experience into playbooks, which describe reusable diagnostic instructions for future incidents with similar faults, and detectors, scripts that take application telemetry as input and output relevant playbook(s). SDO continuously runs the detectors and invokes LLMs only when one fires, so its token cost scales with incident count rather than with operational time. Across 11 open-source microservice applications, SDO cuts resolution time by 52% on recurring faults while keeping per-incident cost low. SDO is open source at https://github.com/self-defining-systems/sdo.

The construction of fault-tolerant atomic registers is a fundamental problem in distributed computing. GPUs, widely used in scientific computing and machine learning, face Byzantine thread behaviour, while their shared memory lacks access control, allowing any thread to modify shared objects. This raises the question of whether atomic register semantics can be maintained when any thread, including the designated writer, may behave Byzantine. We study this problem in the GPU-inspired shared-memory model of Georgiou et al. and present implementations of a Byzantine-linearizable single-writer multi-reader (SWMR) register. CAReg uses constant memory with a quantifiable probability of invalid reads, while SnapObject maintains a bounded window of authenticated writes, eliminating invalid reads and tolerating up to \(n-1\) Byzantine processes through lock-free and wait-free variants. Our results demonstrate how GPU phase-based, lock-step execution can partially compensate for the absence of access control.

Many large enterprise and cloud deployments rely on metrics observability solutions (e.g., Prometheus) to monitor key performance indicators of system health (e.g., CPU, memory, network, app-level statistics). Monitoring and analyzing these metrics presents a significant challenge in terms of cost-latency tradeoffs at scale. Ideally, system operators would like immediately deployable and backwards compatible solutions to improve the cost-latency tradeoffs without significant impact on the fidelity of their observability applications. Unfortunately, existing solutions including state-of-art timeseries databases and custom approximate querying solutions fail on one or more aspects. In this paper, we present the design and implementation of TurboProm, a backwards-compatible drop-in accelerator for metrics observability deployments. TurboProm recasts metrics analytics tasks using a semantic-preserving summary view rather than classical query-centric view. Instead of estimating query results on raw timeseries metrics data, TurboProm uses semantic-preserving summaries that enable high-fidelity analytics with significant cost-latency improvements. We present a practical implementation of TurboProm that can be retrofitted to any existing observability deployment. We evaluate TurboProm on diverse real and synthetic workloads against state-of-art solutions and show that it can provide significant reduction in query latency and cost. TurboProm is open-source at https://github.com/ProjectASAP/ASAPQuery.

Dynamic Tensor Rematerialization (DTR) trains neural networks under a memory budget by evicting tensors and recomputing them on demand. A natural expectation is that a budget large enough to complete training stays sufficient when it is increased. Using DTR's reference simulator, simrd, to replay the public DTR training traces, we show that this expectation fails. We call the exceptions feasibility holes: memory budgets at which an eviction policy runs out of memory although both a smaller and a larger budget complete, a form of non-monotone feasibility. Under a written sweep protocol with dated amendments, a 0.01-step budget grid shows no such hole for DTR on any of six confirmatory traces, although unresolved timeouts prevent it from establishing monotonicity. Finer sampling finds holes on InceptionV4 (three regions) and U-Net (two regions), in addition to one on the ResNet-32 development trace. Every hole is reproduced by three repeats and by unmodified simrd. For one InceptionV4 pair (0.2343 versus 0.235 of peak memory) we trace the failure to individual allocations. The larger budget avoids one eviction at an operator where the smaller budget was 173,479 bytes short. The extra data it carries forward forces a later eviction of a tensor needed three operators afterwards, and rebuilding that tensor triggers a recomputation cascade that cannot fit. Keeping that one tensor resident under the same budget prevents the failure. The two simple score changes we tested do not remove holes across traces; both add failing budgets on some traces. All results are from simulation, with no allocator model; behaviour on GPU hardware is untested.

The scalability and performance of edge federated learning are fundamentally constrained by synchronization bottlenecks. Clients operate under heterogeneous compute, time-varying communication conditions, and, in some settings, privacy constraints, yet synchronization is often still handled with fixed schedules. However, synchronization decisions are often made outside the client, even though the client directly observes local progress, resource state, and communication conditions. We present a client-side reinforcement learning synchronization framework in which each client decides during training whether to upload, download, synchronize bidirectionally, or defer communication. The framework places a lightweight scheduler alongside the local model and formulates synchronization as a resource-aware control problem driven by local learning and system state. We study this framework on two representative edge workloads, spatiotemporal traffic forecasting and image classification. Across both settings, adaptive client-side synchronization improves the accuracy-cost trade-off, reducing system cost in traffic forecasting while maintaining competitive cost and improving final accuracy in image classification. We also consider privacy-constrained settings, where synchronization must account for not only latency and energy but also privacy expenditure. Overall, our results show that synchronization in edge federated learning is better treated as an adaptive client-side control problem than only as a server-side coordination mechanism.

The most accurate agents executing user-provided tasks are driven by large, cloud-hosted models, which solve them by reading the private data of the user. Agents powered by models small enough to run on edge devices keep that data local, but yield far worse results. A third option is to have cloud models write a program that devices execute: the data stays local, but as the program is written before any data is seen, it fails whenever a step depends on interpreting content that appears only at run time. We explore an intermediate architecture, in which the cloud-written program defers bounded questions over user data to a small local model through an ordinary function call. On AppWorld this improves the success rate of a code-only agent in all six combinations of backbone model and test partition we evaluate, by up to 14.9pp, recovering a substantial part of the gap to a cloud-hosted ReAct loop while keeping user data on the device.

Urban traffic systems are becoming increasingly complex, demanding real-time traffic perception, proactive collision prediction, and intelligent management to improve safety and reduce congestion. Although AI-driven traffic management has shown strong promise, its success relies on diverse, high-quality data spanning critical scenarios such as collisions, near-collisions, and traffic rule violations. Digital twins offer a powerful solution by enabling synchronized virtual replicas of physical traffic environments for real-time monitoring, collision prediction, traffic-flow optimization, and diverse realistic scenario generation. However, continuously synchronizing a high-fidelity urban traffic digital twin through edge surveillance cameras requires substantial edge-cloud communication, and transmitting raw video streams leads to severe bandwidth and latency constraints. To overcome this challenge, this paper proposes a semantic communication-based edge-cloud framework for real-time urban traffic digital twin applications. By leveraging edge intelligence to extract and transmit only task-relevant semantic traffic information instead of full video feeds, the proposed framework significantly reduces communication overhead while preserving the essential dynamics needed for accurate real-time digital twin reconstruction. Experimental results demonstrate more than 99% transmission savings and an average trajectory reconstruction error of only 0.12 m, confirming the proposed framework as an effective, scalable, and low-latency solution for AI-enabled urban traffic digital twin applications.