TL;DR
Experimental results on SPEC CPU benchmarks show that MPORA delivers accurate predictions under unseen inputs and distribution shifts with low overhead, while improving schedulability and response times over existing methods.
PRISM, a prediction-guided runtime framework that jointly selects model variants and CPU allocations for containerized edge microservices, and adapts each pipeline stage in place and minimizes predicted CPU-package energy under deadline, resource, and offline model-level Quality of Result constraints is presented.
This paper introduces a job-level dynamic cache allocation strategy for multicore systems in which critical real-time tasks execute on a dedicated core while sharing the LLC with best-effort workloads executing on the remaining cores.
Fine-grained computation--communication overlap in distributed Mixture-of-Experts (MoE) inference allows communication to begin as partial compute results become ready. However, cooperative thread arrays (CTAs) performing computation and communication contend for finite residency capacity on streaming multiprocessors (SMs). Because a resident CTA generally retains its allocated SM resources until completion, CTAs that cannot be co-resident must wait for resources, resulting in wave-like execution. A fixed resource partition cannot adapt to changes in input size, routed expert load, and kernel configuration, potentially causing a communication backlog or reducing expert compute parallelism. We present a wave-quantized analytical model and launch-time resource manager for dependency-coupled overlap pipelines. Using routed-tile counts, kernel occupancy, GPU residency constraints, and split-level readiness dependencies, it selects the communication-CTA count and resource partition before each launch without candidate execution, per-workload profiling, or kernel recompilation. We integrate the method into the public COMET A100 implementation in FLUX. We evaluate three MoE models on four NVIDIA A100 GPUs under several parallelism strategies at the GEMM2+GatherRS operator, complete post-router MoE layer, and complete-model prefill levels. Across 15 real-p90 workloads, the analytical selector achieves 3.22 percent mean regret relative to the measured oracle with a mean solver overhead of 0.157 microseconds. Over COMET, our method achieves geometric-mean speedups of 2.528x at the GEMM2+GatherRS operator, 1.771x at the complete post-router MoE layer, and 1.185x for complete-model prefill, with maxima of 4.218x, 2.584x, and 1.439x, respectively. At every feasible TP=2/EP=2 sequence length of at least 4,096, our implementation outperforms COMET, Megatron core-TE, and FastMoE TP+NCCL.
Power-state management in high-performance computing (HPC) clusters must reduce idle energy without excessive wake-up delays for rigid parallel jobs. This paper presents SNF-ICON, an event-driven controller combining smallest-need-first (SNF) gang scheduling, predictive wake timing, and adaptive warm-spare control. At each scheduler invocation, recent interarrival and completed-service samples are screened for sufficiency, exponential-like variability, low lag-one autocorrelation, and acceptable Kolmogorov-Smirnov distance. Rejected or data-sparse windows use SNF+IPM (Intelligent Power Manager), whereas accepted windows activate release prediction and an exponential next-event model. Warm-spare optimization is applied only when queue, event, and arrival-recency conditions permit, balancing estimated waiting and non-compute energy over a timeout-capped horizon. We evaluate four DAS2 trace segments and a generated Markovian workload on AOBA-derived 64-node models, plus SDSC Blue on an AOBA-derived 1152-node model. SNF-ICON is compared with SNF+IPM and First Come First Served (FCFS) + backfilling with IPM. It reduces average waiting time relative to the FCFS-based baseline in all six cases and remains close to at least one heuristic energy baseline in five. The generated workload spends substantial time in ICON mode, whereas DAS2 workloads operate mainly in fallback. Furthermore, cross-platform results show strong dependence on node-transition and power models. Thus, no single policy or parameter set works best in every case.
Apache Spark is widely used for distributed data processing, but accurately predicting application execution time remains challenging because performance depends on application structure, resource configuration, and executor‐allocation behavior. This article presents two deterministic, graph‐based simulation models for predicting the execution time of Spark applications under static and dynamic executor allocation. The first model, the Static Allocation Model (SAM), represents an application as a monolithic directed acyclic graph (DAG) of stages and simulates execution using Spark's scheduling behavior under a fixed number of executor cores. To capture more realistic execution environments, this framework is extended to a Dynamic Allocation Model (DAM), which accounts for runtime changes in executor resources using backlog‐time and idle‐time thresholds. DAM dynamically adjusts executor allocation during execution and integrates these changes within the deterministic scheduling framework. Validation of SAM on three TPC‐DS benchmark queries—Query‐64, Query‐26, and Query‐52—yielded prediction errors of 2.9%, 4.3%, and 0.4%, respectively, with a mean absolute error of 2.5%. Experimental evaluation of DAM on TPC‐DS Query‐52 and the SparkBench K‐Means workload yielded prediction errors of 5% and 4.7%, respectively. To the best of our knowledge, this work presents the first deterministic approach for modeling Spark application execution time under dynamic executor allocation. The proposed models are lightweight, interpretable, and require minimal input data, offering a practical alternative for performance prediction and resource optimization in distributed data‐processing systems.
We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.
