Mahesh Subramaniam, Director of Product Management, Data Center, Juniper Networks

Juniper Networks AI Data Center and Ultra Ethernet Consortium (UEC)

AI & MLData Center
Mahesh Subramaniam Headshot
The opening sequence shows the host, Mahesh Subramaniam, in front of a green screen that includes the Juniper logo. The onscreen text reads, “Mahesh Subramaniam, Director of Product Management, Data Center”.

Paving the way to a complete, open communications-stack architecture.

Juniper has joined the Ultra Ethernet Consortium (UEC) as part of its dedication to constructing open, high-performance solutions with improved economics to address the increasing network demands of AI data centers. Explore the video to understand the benefits and advantages of delivering a full-communications stack architecture for constructing AI-optimized data center networks that simplify our customers' experiences.

Show more

You’ll learn

  • How Juniper will participate in technical workgroups to address vital requirements in next-generation networks

  • Why job completion time (JCT) is a fundamental concept for emerging data center infrastructure requirements and what it will take to reduce JCT

Who is this for?

Network Professionals


Mahesh Subramaniam Headshot
Mahesh Subramaniam
Director of Product Management, Data Center, Juniper Networks


0:06 I'm happy to share that Juniper is now a

0:09 member of ultra ethernet Consortium ueec

0:13 and we are participating in all the

0:15 active technical work groups in U but

0:18 what is U ueec is a Consortium of

0:21 leading vendor working together to build

0:24 a common architecture that helps to

0:27 scale and optimize the ethernet fabrics

0:29 for for AI and HPC infrastructures as

0:32 you all know our existing ethernet

0:35 technology has demonstrated its

0:36 reliability numerous times over the past

0:39 few decades several technologies have

0:42 challenged it over the years but

0:44 ethernet has continued to dominate and

0:47 forers a powerful common vendor

0:49 ecosystem that drives strong multivendor

0:52 and strong interoperable technologies

0:54 that technology consistently evolves to

0:57 address the demands of future networks

0:59 now now the U focus on a high

1:02 performance High scalable common stock

1:04 architecture to meet the growing Network

1:06 demands specifically on AI and HBC

1:11 Fabrics why is Juniper fought of

1:14 U machine learning infrastructure

1:16 requirements are expanding rapidly and

1:19 anticipated to grow for the foreseeable

1:21 features so Juniper will help resolve

1:24 networking issues paused by AI ML

1:27 workloads and address the technical

1:30 challenges like low entropy elephant

1:32 flows that Traverse many protocol layers

1:36 as a member of all current technical

1:37 work groups Juniper will participate in

1:40 driving all the architecture requirement

1:42 to solve the broader and more difficult

1:44 customer design requirements plus our

1:47 long history of building low touch High

1:50 automated D Center Fabric and high scale

1:53 automated Network management will bring

1:55 value to U one point one point I would

1:59 like like to make it clear whenever we

2:02 talk about aidc infrastructure

2:04 requirements a key word that we cannot

2:07 avoid is JCT job completion time in

2:10 order to reduce the job completion time

2:13 on training cluster we need to ensure

2:16 low tail latency on the fabric to ensure

2:19 the low tail latency on the fabric there

2:21 are few vital requirements from U which

2:24 needs to be focused for the Next

2:26 Generation networks such as multipathing

2:29 and packet spraying flexible delivery

2:32 order modern congestion and control

2:34 mechanism end to end Telemetry scale

2:37 stability and relability of course as of

2:40 today Juniper has a powerful

2:43 comprehensive data center switching

2:44 portfolio on the hardware side we have a

2:47 high Ric 400 gig and upcoming 800 gig

2:50 switches to accommodate the GPU cluster

2:52 interconnectivity on the software side

2:55 an efficient load balancing solution for

2:57 packet spray Rocky V2 and dcq to control

3:01 the congestion on the fabric at the top

3:05 Juniper has abstra the only multivendor

3:08 DC fabric management and automation tool

3:11 in the industry which is great at

3:13 operating data centers as well upcoming

3:16 days collaboration with u we are sure to

3:19 develop many other hardware and software

3:22 capabilities that will lead to the

3:23 better and better AI ml infrastructure

3:27 for data center operators and better AI

3:31 experience for inducers like

3:41 you

Show more