The storage network is the network used by the block storage services on the host, enabling data exchange between nodes within the ACOS cluster. Each node in an ACOS cluster is allowed to have only one storage network.
The main factors affecting storage network performance are packet loss, network latency, and network interruption. Their respective impacts are described below.
When a single node experiences frequent packet loss due to issues such as a faulty NIC, the storage service on that node will experience a significant increase in latency. When the packet loss rate reaches or exceeds 10%, maintaining stable connections becomes difficult. The network can be considered essentially unavailable, and the node is likely to enter an overall abnormal state. At the same time, configuration change requests in the ACOS cluster—including the creation, deletion, update, or other operations of storage objects—will experience significant delays.
When the storage network latency is between 5 ms and 500 ms, it does not affect cluster availability. However, the higher the network latency, the lower the cluster's performance. Latencies above 500 ms may lead to service instability.
When network latency exceeds 1 second (the ZooKeeper election timeout threshold), the ZooKeeper Leader cannot be successfully elected, and the ACOS cluster will become unavailable.
When the storage network of a single node is interrupted, alarms for node state anomaly and node unreachable will be triggered.
The impact of single-node or multi-node storage network interruptions is equivalent to node failures. For details, see Node failures.
