Pith. sign in

REVIEW 6 cited by

An Extensible Software Transport Layer for GPU Networking

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2504.17307 v2 pith:GZJDTBHQ submitted 2025-04-24 cs.NI

An Extensible Software Transport Layer for GPU Networking

classification cs.NI
keywords transportrdmanetworkingnicssoftwareucclworkloadscollisions
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Fast-evolving machine learning (ML) workloads have increasing requirements for networking. However, host network transport on RDMA NICs is hard to evolve, causing problems for ML workloads. For example, single-path RDMA traffic is prone to flow collisions that severely degrade collective communication performance. We present UCCL, an extensible software transport layer to evolve GPU networking. UCCL decouples the data path and control path of existing RDMA NICs and efficiently runs the control-path transport on host CPUs. This software extensibility brings in transport innovations that cannot be achieved in hardware for ML workloads, e.g., a multipath transport to resolve flow collisions. ML collectives atop UCCL achieve up to 4.5x higher performance compared to existing RDMA NICs.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. OrderMoE: An expert similarity driven distributed edge MoE inference

    cs.NI 2026-07 conditional novelty 6.0

    OrderMoE reduces cross-server traffic in edge MoE inference by substituting a similar local expert for the router's remote target expert, with only small quality loss.

  2. GF-DiT: Scheduling Parallelism for Diffusion Transformer Serving

    cs.DC 2026-06 unverdicted novelty 6.0

    GF-DiT introduces elastic GPU parallelism scheduling for DiT serving via asynchronous trajectory tasks and group-free collectives, reporting up to 6.01x throughput gains over static configurations.

  3. GF-DiT: Scheduling Parallelism for Diffusion Transformer Serving

    cs.DC 2026-06 unverdicted novelty 6.0

    GF-DiT dynamically adapts parallelism during DiT serving via trajectory tasks and group-free collectives, reporting up to 6x throughput and 95% latency reduction versus static configurations.

  4. UCCL-Zip: Lossless Compression Supercharged GPU Communication

    cs.DC 2026-04 unverdicted novelty 6.0

    UCCL-Zip adds lossless compression to GPU communication to reduce LLM bottlenecks while preserving exact numerical correctness.

  5. TENT: A Declarative Slice Spraying Engine for Performant and Resilient Data Movement in Disaggregated LLM Serving

    cs.DC 2026-04 conditional novelty 6.0

    A telemetry-driven slice-spraying transfer engine with late-binding path selection improves LLM serving throughput and self-heals from link failures in tens of milliseconds.

  6. PICO: Performance Insights for Collective Operations

    cs.DC 2025-08 unverdicted novelty 6.0

    PICO is a benchmarking framework for collective operations that decouples portable setup from platform execution, supplies reference MPI implementations, and shows default choices can be up to 5x slower with up to 44%...