Data Center Network

Challenge

Large-scale data centers are growing rapidly, evolving towards expansive, continuous, and scalable deployments. The construction of DCNs presents new challenges and requirements.

Separated computing and storage

•Computing nodes and storage nodes are deployed separately, addressing the mismatch in hardware requirements for different resources. •The separated architecture simplifies the cluster planning and budget, allows for independent scheduling of computing resources, and reduces costs associated with rapid scalability. •Interaction between storage and computing nodes significantly increases east-west traffic.

Distributed computing for machine learning

•Machine learning requires data shuffling and parameter synchronization, and data needs to be processed in blocks due to limited local capacity of a single node. •Multiple computing nodes need to randomly exchange data blocks. •The reduce algorithm is required for mesh computation during parameter update, and logical loop traffic exists between computing nodes.

Automated O&M

•Thousands of devices in a data center need to be deployed in a short period of time. ZTP meets urgent service requirements. •Repetitive and routine tasks can be automated, reducing the need for manual intervention.

Fine-grained management

•IT operation personnel focuses on reducing the packet loss rate, managing and preventing network congestion, controlling network traffic, setting packet priorities, and more. •Network traffic distribution, service applications, and network status are visualized through intuitive analysis charts and reports.

Personalized innovations

•IT operation personnel focuses on reducing the packet loss rate, managing and preventing network congestion, controlling network traffic, setting packet priorities, and more. •Network traffic distribution, service applications, and network status are visualized through intuitive analysis charts and reports.

Overview of the Solution

Large-Scale
Networking

High
Performance

High
Availability

Easy
O&M

Networking using single-chip fixed switches and multi-plane horizontal scaling, supporting networking of tens of thousands of servers

Single-chip high-bandwidth design to achieve 200G access to a single server and forwarding latency in microseconds

The entire network uses only the BGP protocol, which simplifies maintenance and facilitates expansion.
BGP is optimized to shorten the failover time to seconds, providing ultimate high availability.
Layer 3 IP forwarding is used on the entire network, and ECMP paths are formed through BGP, ensuring normal service operations upon single failures and delivering high availability.

Obtaining real-time status data such as cache and queue data based on gRPC
Subscribing to key information based on gRPC and reporting it periodically
Obtaining key state packets based on ERSPAN

Feature Highlight

Large-Scale Networking

The solution supports up to 3k servers per PoD and up to 19k servers per cluster. Hybrid networking with 25G and 100G is supported, ensuring seamless transition from 25G to 100G.

High Performance

High-performance chips enable 200G server access to a single server, while single-chip designs ensure low-latency forwarding.

High Availability

De-stacking technology improves the stability, ECMP paths ensure high availability, and the lossless network reduces the packet loss rate.

Easy O&M

Devices go online automatically, standardized configuration is delivered, visual monitoring is implemented, and fault information is collected with one click.

Data Center Network Solution

Challenge

Overview of the Solution

Feature Highlight

Large-Scale Networking

High Performance

High Availability

Easy O&M

Related Products

M2-W6940-128QC

M2-W6510-48GT4V

M2-W6510-32C

M2-W6930-64QC

M2-S6920-32QC2X

M2-S6930-64QC

M2-S6520-24DC8QC

M2-S6510-32C

M2-S6920-4S

M2-S6510-48V8C

M2-S6510-48X8C

M2-S6000-48GT4X

M2-W6920-32QC2X

M2-W6520-24DC8QC

M2-W6510-48V8C

M2-W6940-64OC

Request More Information

Preferences