Remove Bandwidth Remove Network Remove Topology
article thumbnail

A RoCE network for distributed AI training at scale

Engineering at Meta

AI networks play an important role in interconnecting tens of thousands of GPUs together, forming the foundational infrastructure for training, enabling large models with hundreds of billions of parameters such as LLAMA 3.1 Distributed training, in particular, imposes the most significant strain on data center networking infrastructure.

Network 132
article thumbnail

Using Chakra execution traces for benchmarking and network performance optimization

Engineering at Meta

Meta presents Chakra execution traces , an open graph-based representation of AI/ML workload execution, laying the foundation for benchmarking and network performance optimization. At Meta, our endeavors are not only geared towards pushing the boundaries of AI/ML but also towards optimizing the vast networks that enable these computations.

Network 109
Insiders

Sign Up for our Newsletter

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.

Trending Sources

article thumbnail

Practical Steps for Enhancing Reliability in Cloud Networks - Part I

Kentik

When evaluating solutions, whether to internal problems or those of our customers, I like to keep the core metrics fairly simple: will this reduce costs, increase performance, or improve the network’s reliability? It’s often taken for granted by network specialists that there is a trade-off among these three facets. Durability.

Cloud 104
article thumbnail

How Meta trains large language models at scale

Engineering at Meta

Supporting GenAI at scale has meant rethinking how our software, hardware, and network infrastructure come together. Solving this problem requires a robust and high-speed network infrastructure as well as efficient data transfer protocols and algorithms. requires revisiting trade-offs made for other types of workloads.

article thumbnail

Network observability: Hype or reality?

Kentik

If you haven’t yet heard the term “network observability,” you will be hearing it soon. Some say that network observability is just marketing hype from vendors. They say, “networks have always been observable, so there’s nothing new here.” I say network observability is not just vendor hype, and this blog will make the case.

Network 85
article thumbnail

Securing Your Network Against Attacks: Prevent, Detect, and Mitigate Cyberthreats

Kentik

As networks become distributed and virtualized, the points at which they can be made vulnerable, or their threat surface , expands dramatically. This is compounded by recent trends of remote work, where network operators need to wrestle with the fact that employees often access the network via work sites with far less governance.

Network 94
article thumbnail

Why latency is the new outage

Kentik

Latency is not a new problem in networking. Not as difficult as time travel, but it’s difficult enough so that for 30+ years IT professionals have tried to skirt the issue by adding more bandwidth between locations or by rolling out faster routers and switches. Fast forward to today, most networks enjoy five nines (99.999%) of uptime.

TCP 116