{"slug": "fleetsieve-decision-critical-profiling-for-slo-aware-llm-fleet-configuration", "title": "FleetSieve: Decision-Critical Profiling for SLO-Aware LLM Fleet Configuration", "summary": "FleetSieve, a new profiling method for LLM serving fleets, reaches the oracle aggregate decision using 22,200 GPU-seconds, 6.9% less than uniform random profiling on a fixed H100 grid for a 31B-parameter model. Across 200 random reveal orders, its mean saving is 5.4% (95% CI: 3.5-7.2%), and it avoids SLO violations by modeling capacity and tail latency jointly.", "body_md": "arXiv:2608.19659v1 Announce Type: new\nAbstract: Choosing tensor-parallel (TP) degrees and replica counts for an LLM serving fleet is difficult because performance is not monotonic in TP and the feasible choice can change with load. Exhaustive profiling resolves this uncertainty, but measures many configurations that do not affect the final resource allocation. We present FleetSieve, which selects measurements according to their expected effect on a resource-coupled, SLO-aware fleet decision. FleetSieve models capacity and tail latency jointly, compares conservative and optimistic allocations, and stops when their remaining decision gap is below a specified tolerance. On a fixed H100 measurement grid for a 31B-parameter open-weight model, FleetSieve reaches the oracle aggregate decision using 22,200 GPU-seconds, 6.9% less than uniform random profiling in the fixed comparison. Across 200 random reveal orders, its mean saving over random profiling is 5.4% (95% bootstrap CI: 3.5-7.2%). The fixed-comparison saving is 21.5% for Chat, while FleetSieve does not use the fewest GPU-seconds for Code. Joint capacity and tail modeling also avoids selecting a configuration whose 46.4-second completion p99 violates a 30-second SLO. In a 16-GPU allocation, an incorrect sparse-profile decision loses up to 1.93 requests/s and 12.4 percentage points of max-min fulfillment. Boundary repeats and BurstGPT measurements support the observed load-dependent tail-latency mechanism.", "url": "https://wpnews.pro/news/fleetsieve-decision-critical-profiling-for-slo-aware-llm-fleet-configuration", "canonical_source": "https://www.machinebrief.com/news/fleetsieve-decision-critical-profiling-for-slo-aware-llm-fle-clc7", "published_at": "2026-08-21 04:00:00+00:00", "updated_at": "2026-08-21 04:14:26.388453+00:00", "lang": "en", "topics": ["machine-learning", "ai-infrastructure", "ai-research"], "entities": ["FleetSieve", "arXiv", "H100", "BurstGPT"], "alternates": {"html": "https://wpnews.pro/news/fleetsieve-decision-critical-profiling-for-slo-aware-llm-fleet-configuration", "markdown": "https://wpnews.pro/news/fleetsieve-decision-critical-profiling-for-slo-aware-llm-fleet-configuration.md", "text": "https://wpnews.pro/news/fleetsieve-decision-critical-profiling-for-slo-aware-llm-fleet-configuration.txt", "jsonld": "https://wpnews.pro/news/fleetsieve-decision-critical-profiling-for-slo-aware-llm-fleet-configuration.jsonld"}}