High-performance Computing (HPC) Networking
Cisco DNA / Arista CloudVision® for H100 InfiniBand Solution
Overview
The solution based on the NVIDIA® H100 GPU, integrated with Cisco DNA or Arista CloudVision® management platforms, offers a robust Infiniband network architecture tailored for high-performance computing (HPC). This setup meets various business needs by encompassing the InfiniBand network, management network, and storage network.
InfiniBand
InfiniBand is a high-performance networking technology primarily used in HPC environments, data centers, and enterprise networks.
Purpose: InfiniBand is designed to deliver high-speed data transfer, low latency, and high throughput for communication between servers and storage systems.
In-band Management
In-band management involves using the same network that handles regular data traffic to manage network devices and systems.
Purpose: This method allows administrators to perform tasks such as configuration, monitoring, and troubleshooting over the same network used for normal data communication.
Out-of-band Management
Out-of-band management employs a dedicated management network separate from the regular data network for managing network devices and systems.
Purpose: OOB management provides a separate pathway for management tasks, ensuring that administrators can access network devices even if the primary data network is down.
InfiniBand Network Components
Powered by NVIDIA® H100 GPUs and InfiniBand switches, the InfiniBand network offers ultra-low latency and high bandwidth. It ensures lossless transmission through advanced flow control mechanisms.
| S.No | Product Equivalents |
|---|---|
| 1 | Cisco: Nexus N9K-C9364C-GX, Nexus 92348GC-X Arista: 7800 R3 Series |
| 2 | Cisco: QSFP-DD-800G-SR8 Arista: QSFP-DD-800G-SR8 NXTGEN: OSFP (800G) SR8 Transceiver |
| 3 | Cisco: QSFP-400G-SR4-S Arista: OSFP-400G-SR4 NXTGEN: QSFP-DD (400G) FR4 Transceiver |
Management Network
Cisco and Arista switches leverage the advanced capabilities of the Cisco DNA or Arista CloudVision management platforms. These platforms enable customers to efficiently provision, monitor, manage, proactively troubleshoot, and maintain their HPC infrastructure, resulting in higher utilization and reduced overall operational expenses.
| S.No | Product Equivalents |
|---|---|
| 1 | Cisco Nexus 93240YC-FX2, Arista 7050SX3-48YC12 |
| 2 | Cisco WS-C3850-48XS-F-S, Arista 7048T-A |
| 3 | Cisco Nexus 9364C, Arista 7060CX-64 |
| 4 | Cisco Nexus 93180YC-EX, Arista 7280SR-48C6 |
| 5 | NXTGEN: 10.3Gbps SFP+ Transceiver NXTGEN: 25Gbps SFP28 Transceiver |
| 6 | NXTGEN: QSFP28 SR4 Transceiver |
Storage Network
CISCO/ARISTA switches support the BGP protocol with powerful routing control capabilities while ensuring the optimal forwarding path and low-latency forwarding status of the storage network. They are flexible and can scale to meet specific capacity and bandwidth requirements.
| S.No | Product Equivalents |
|---|---|
| 1 | Cisco Nexus 9364C, Arista 7060CX-64 |
| 2 | NXTGEN: QSFP28 SR4 Transceiver |
RoCE Computing
Enhance HPC Network with a 400G RoCE Lossless Solution
This solution provides an optimal 400G interconnect for QSFP-DD switches and OSFP network cards, resolving compatibility issues between different port encapsulations. It is designed for the network topology of HPC architectures, encompassing RoCE compute networks, management networks, and storage networks, to meet diverse business requirements.
| S.No | Product Equivalents |
|---|---|
| 1 | Cisco Nexus 9364C, Arista 7060CX-64 |
| 2 | Cisco Nexus 93240YC-FX2, Arista 7050SX3-48YC12 |
| 3 | Cisco Nexus 9332C, Arista 7060CX-32S |
| 4 | Cisco Nexus 9364D-GX2A, Arista 7060DX5-64S |
| 5 | NXTGEN QSFP28 SR4 transceiver 100G |
| 6 | NXTGEN 25Gbps SFP28 Transceiver |