arXiv subjects

Matty Kadosh

Publications and source records attributed to Matty Kadosh.

4 recordsLinked to original sources

arXiv2026

High-speed Networking for Giga-Scale AI Factories

As distributed model training scales to span hundreds of thousands of GPUs, scale-out networks face unprecedented performance and efficiency demands. NVIDIA Spectrum-X Ethernet has been designed from the ground up to achieve predictable and stable network performance with high utilization and low latency. This paper presents the Spectrum-X multiplane architecture, which replaces hierarchical depth with topological parallelism, and introduces hardware-accelerated load balancing in NICs and switches as the key architectural approach to provide fast reaction to highly dynamic network conditions at the microsecond timescales that AI training workloads demand. We describe the motivation, design principles, evaluation methodology and performance on state-of-the-art benchmarks, as well as the lessons we learned from deploying and debugging Spectrum-X networks in large-scale systems. Our evaluation highlights production-grade AI infrastructure performance across three core dimensions: 98% of the theoretical line rate with low jitter-free latency; strong cross-tenant isolation for concurrent workloads; robust, capacity-proportional bisection bandwidth and 7% latency increase for 10% fabric link failures; and rapid reaction to host and fabric link flaps during LLM training workloads.

cs.NI↗

arXiv2026

Avoiding Cross-Datacenter Collective Congestion via Disaggregated Buffering

LLM training at the scale of tens of thousands of GPUs now spans multiple datacenters (DC), making cross-DC collectives over long-haul links unavoidable. A critical and overlooked bottleneck arises when these collectives collide with intra-DC traffic at the destination - a common pattern in real workloads. The multi-millisecond congestion control loop is too slow to react, triggering severe packet loss and congestion collapse. We present Spillway, a transparent in-network mechanism that buffers dropped packets in switch-disaggregated buffers in a destination data center and drains them once congestion subsides. Through large-scale end-to-end simulations and a hardware prototype, we show that Spillway eliminates performance degradation from collective collisions, reducing iteration time by up to 14 %, without changes to end hosts or training frameworks.

cs.NI↗

arXiv2023

OptiReduce: Resilient and Tail-Optimal AllReduce for Distributed Deep Learning in the Cloud

We present OptiReduce, a new collective-communication system for the cloud with bounded, predictable completion times for deep-learning jobs in the presence of varying computation (stragglers) and communication (congestion and gradient drops) variabilities. OptiReduce exploits the inherent resiliency and the stochastic nature of distributed deep-learning (DDL) training and fine-tuning to work with approximated (or lost) gradients -- providing an efficient balance between (tail) performance and the resulting accuracy of the trained models. Exploiting this domain-specific characteristic of DDL, OptiReduce introduces (1) mechanisms (e.g., unreliable bounded transport with adaptive timeout) to improve the DDL jobs' tail execution time, and (2) strategies (e.g., Transpose AllReduce and Hadamard Transform) to mitigate the impact of gradient drops on model accuracy. Our evaluation shows that OptiReduce achieves 70% and 30% faster time-to-accuracy (TTA), on average, when operating in shared, cloud environments (e.g., CloudLab) compared to Gloo and NCCL, respectively.

cs.DC↗

arXiv2020

LB Scalability: Achieving the Right Balance Between Being Stateful and Stateless

A high performance Layer-4 load balancer (LB) is one of the most important components of a cloud service infrastructure. Such an LB uses network and transport layer information for deciding how to distribute client requests across a group of servers. A crucial requirement for a stateful LB is per connection consistency (PCC); namely, that all the packets of the same connection will be forwarded to the same server, as long as the server is alive, even if the pool of servers or the assignment function changes. The challenge is in designing a high throughput, low latency solution that is also scalable. This paper proposes a highly scalable LB, called Prism, implemented using a programmable switch ASIC. As far as we know, Prism is the first reported LB that can process millions of connections per second and hundreds of millions connections in total, while ensuring PCC. This is due to the fact that Prism forwards all the packets in hardware, even during server pool changes, while avoiding the need to maintain a hardware state per every active connection. We implemented a prototype of the proposed architecture and showed that Prism can scale to 100 million simultaneous connections, and can accommodate more than one pool update per second.

cs.NI↗