Many cluster failures are traced back to the same reason: management, live migration, backup and business traffic are squeezed into the same uplink. Normally everything is fine, but a backup or migration can starve the management network to death, causing the host to misjudge and lose connection. The solution is to separate the channels.
Four types of traffic and two division principles
- Management, vMotion or live migration, storage, and services each use independent VLANs, and the management network has fixed reserved bandwidth.
- Sudden large traffic (migration, backup) occupies a separate physical channel and does not interfere with steady-state storage traffic.
- When conditions permit, the storage network is further subdivided into two layers: front-end business access and back-end replication synchronization.
How to divide the limited network card into ports
When budget is tight, prioritize storage and migration channels.PCI-EX8 25G dual optical port fiber optic network card (Intel E810 chip)The two optical ports are connected to two storage switches to form redundancy; for business and management sidePCI-EX4 10 Gigabit single electrical port network card (Intel X550 chip)Just reuse existing copper wiring. A two-node small cluster would rather have less aggregation, but also ensure isolation first.
The details of aggregation and fault tolerance
It is recommended to use LACP dynamic aggregation for multiple ports with similar traffic, so that services will not be lost due to a single line failure; live migration is generally not suitable for pass-through, and it is more flexible to place it on a virtual switch; the number, gateway, and MTU recommendation list of each VLAN should be kept on file. Line-by-line verification during troubleshooting is much faster than temporary trial and error.
Channel division and network card quantity calculation vary greatly due to environmental differences. Please call 0755-83179002 to check with EB-LINK technical support before purchasing.