Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.24328

SigmaTrace: Library-Agnostic Observability for CPU/GPU-Initiated RDMA Communication

Javid Baydamirli·Didem Unat
View original source ↗Read source document ↗Export metadata ↓

Abstract

RDMA networking enables low-latency, high-bandwidth communication on large systems. Despite its adoption, however, observability is lacking. Existing tools target specific communication libraries or only report aggregated data, missing per-operation behavior. The gap is widest for GPU-initiated networking, whose in-kernel RDMA operations are invisible to the host. We present SigmaTrace, a library-agnostic RDMA profiling framework providing fine-grained, per-operation visibility for CPU and GPU-initiated networking. It traces operations at the libibverbs/mlx5dv layers, the common denominator beneath communication libraries on InfiniBand and RoCE, and employs a low-overhead memory interposition technique that, to our knowledge, gives the first observability into GPU-initiated RDMA without source patching or binary instrumentation. Host-side tracing adds 1.3% overhead on NCCL AllReduce at 128 GPUs and under 2% on RCCL at 1024 GPUs. On a persistent-kernel BFS and Mixture-of-Experts inference, SigmaTrace exposes an in-kernel barrier spread and 2.13x inter-node traffic skew invisible to existing profilers.

Explore related subjects

cs.DC

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Javid Baydamirli, Didem Unat. 2026-10-16. SigmaTrace: Library-Agnostic Observability for CPU/GPU-Initiated RDMA Communication. https://arxiv.org/abs/2610.24328

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Load Balancing with Evolving Workloads in Large-Scale Language Model Serving

Serving large language models requires assigning requests whose workloads grow during processing and whose completion times are unknown. With a synchronization barrier, lightly loaded workers wait for the slowest worker at every decode iteration. We model this loss through aggregate workload imbalance and formulate a joint admission and assignment problem with sticky requests and capacity limits. Short lookahead incorporates predicted completions of ongoing requests. For exact optimization without lookahead, we prove a comparison under independent geometric decode lengths, bounded independent nondegenerate prompt sizes, and batched releases satisfying a cumulative supply condition. The long-run expected imbalance improvement over first come, first served admission with request-count join-shortest-queue grows at least as the square root of the product of worker capacity and the logarithm of worker count, in the stated scaling regime. The comparison extends to capacity-aware round robin, random routing, and power of two choices. BalanceRoute approximations connect the formulation to practical routing. Experiments using up to 144 NPUs with proprietary and public traces show lower workload spread and higher throughput than the evaluated baselines. These empirical results complement the guarantee for exact optimization.

cs.DC↗
arXiv2026

Coral: Cost-Efficient Multi-LLM Serving over Heterogeneous Cloud GPUs

The landscape of large language models (LLMs) has become increasingly fragmented, with no single model dominating the field. Meanwhile, cloud providers offer a diverse array of mid-tier and older-generation GPUs. These heterogeneous resources provide better availability and deliver comparable performance-per-dollar to top-tier hardware. To efficiently harness these GPUs for concurrent multi-LLM serving, we introduce Coral: an adaptive, heterogeneity-aware serving system. The key idea behind Coral is the joint optimization of resource allocation and replica serving strategies across all models. To keep pace with shifting throughput demands and resource availability, Coral employs a lossless, two-stage decomposition. This approach preserves joint optimality while reducing online solve times from hours to mere tens of seconds. Evaluations across 6 models and 20 GPU configurations demonstrate that Coral reduces serving costs by up to 2.79$\times$ compared to the strongest baseline, while delivering up to 2.39$\times$ higher goodput under scarce resource constraints.

cs.DC↗
arXiv2026

Domain Extension of Lock-Freedom and Wait-Freedom for Group Computations

A domain extension of a definition refers to broadening the scope of a definition so that it applies to a larger set of cases than originally specified. The notion of lock-free and wait-free computation is designed for the domain of tasks that are completed by a single thread (in competition with other threads). The goal of this paper is to extend the definition of lock-freedom and wait-freedom to group-computations (denoted by gl-freedom and gw-freedom) that require that the task at hand must be completed by a collaboration between multiple threads. When extending a definition, certain constraints must be respected: the new domain must remain logically consistent with the original meaning, the extension should not introduce contradictions or ambiguities, and it must preserve the essential properties that make the definition valid and useful. We demonstrate this by showing that our extended definition is consistent with the original definition when the group consists of a single thread. We note that extension allows us to characterize programs in a new domain (distributed computing, NUMA computation systems, systems with private data for different threads, etc.) instead of relegating them to be in the same category (deadlock/livelock-free) without regard to the actual properties of that program. We also illustrate this definition with various examples.

cs.DC↗