Recently, I communicated with a financial user. The customer said that the current application containerized transformation requires tens of millions of small file logs, configuration files, shared files, etc. to be stored, containers can only be stored in local file systems. Thousands of containers need to be shared across containers, which leads to the need for persistent data sharing and exchange. At the same time, the production center and the local disaster recovery center can perform dual-active disaster recovery, when the production center fails, the container client can automatically switch to the local disaster recovery file storage, hoping to realize the dual-active switching of the local three-tier network. Currently, file storage needs to be manually mounted, which is cumbersome to operate. We hope to improve the optimized architecture.

Indeed, with the acceleration of the transformation of stateful containers, the need to share data among containers is relatively wide. However, the original NFS client of the OS does not have sufficient FailOver capability and the FailOver is slow. Therefore, manual FailOver and Mount are required. Based on the common needs of financial users, Huawei storage pioneered the development of NFS Client software plug-ins, similar to multi-path software of SAN storage. Multiple virtual VIPs for NAS file storage can be configured on NFS clients, when one VIP1 is inaccessible, you can switch to another VIP2 through the NFS client, which enables automatic and fast switching. Combined with the NAS file storage dual-active capability, dual-active switching between production and the same city can be realized. The principle of the architecture diagram and the switching effect are shown in the following figure. It can be seen that the switching time is about 14 seconds for the layer -2 network and 20 seconds for the layer -3 network.

In addition, based on the multi-path concurrent balancing capability of NFS clients, I/O is distributed across multiple links at the same time. Compared with the original client, I/O is distributed across a single link, which basically improves the performance throughput by more than four times. In addition, NFS over ROCE is supported to further improve the performance of enterprise NAS file system networks and reduce I/O latency.

For bare metal BMS and virtual machine ECS scenarios, NFS client multi-path capability is also supported.

Source: Bi xunshuo