This topic describes the specifications of Tair (Enterprise Edition) persistent memory-optimized instances. These specifications include the memory capacity, maximum number of connections, and maximum bandwidth.
Instance specifications
The following tables list the specifications of individual shards. The overall performance of a cluster or read/write splitting instance can be calculated by using the following formula: Overall performance = Performance of a single shard × Number of shards.
Note
The Enterprise SSDs (ESSDs) listed in these tables are provided for dedicated storage of system files, such as logs, backups of temporary files, and append-only files (AOFs). They are not intended for data storage.
For information about the differences between the direct connection mode and the proxy mode that are supported by cluster instances, see Cluster architecture.
Overall performance of an instance = Performance of a shard × Number of shards.
Note
When a cluster instance that runs in proxy mode is accessed, the maximum bandwidth is 2,048 Mbit/s, the maximum total QPS is 10,000,000, and the maximum number of connections is 500,000.
For example, assume that an instance has four shards of the tair.scm.with.proxy.standard.2m.8d type. Each shard has the following specifications:
Number of CPU cores: 3
Bandwidth: 96 Mbit/s
Maximum number of connections: 10,000
In this case, the instance provides the following performance:
Number of CPU cores: 12
Bandwidth: 384 Mbit/s
Maximum number of connections: 40,000
Notes about CPU cores
To ensure service stability, the system reserves a CPU core to process
background tasks.
In a cluster instance or a read/write splitting instance, the system reserves a CPU core for each shard or read replica to process background tasks.
Bandwidth description
Each bandwidth value in the preceding tables represents the maximum bandwidth of a single shard. The total bandwidth of a cluster or read/write splitting instance is the combined bandwidth of all shards in the instance.
The maximum bandwidth of a cluster instance in proxy mode or a read/write splitting instance is 2,048 Mbit/s. After the upper limit is reached, the bandwidth cannot be increased even if you add more shards to the instance. Cluster instances in direct connection mode have no limits on the overall bandwidth.
The bandwidth applies to the upstream and downstream bandwidths. For example, if the overall bandwidth of an instance is 24 Mbit/s, the upstream and downstream bandwidths of the instance are both 24 Mbit/s.
The bandwidth limits of Tair and Redis Open-Source Edition refer to the data transfer capacity of individual shards within their respective distributed systems. The limits are generally independent of the network connection types used by clients to connect to the shards.
Proxy mode: 500,000. If the upper limit of 500,000 connections is reached, subsequent connections cannot be established even if you add more shards or nodes.
Direct connection mode: The maximum number of connections to a single shard is 10,000. The maximum number of connections to an instance can be calculated by using the following formula: Number of shards × 10,000.
Maximum number of new connections per second
The maximum number of new connections per second refers to the number of connections that can be established per second. For example, assume that the maximum number of new connections to an instance per second is 10,000, the maximum allowed number of connections to the instance is 50,000, and the actual number of connections at the Nth second after the instance starts to run is 12,000. In this case, the maximum number of connections at the (N+1)th second is 22,000. This value is calculated by using the following formula: 12,000 + 10,000.
Note
If the instance is connected in direct connection mode, connection performance degrades because proxy nodes are bypassed. The maximum number of new connections to a single data shard per second is 2,000. For example, if an instance has four data shards, the maximum number of new connections to the instance per second is 8,000.
Elastic Compute Service (ECS)
Lingma
