Specifications for memory-optimized Tair (Enterprise Edition) instances with a read/write splitting architecture, including read replica count, memory capacity, maximum connections, bandwidth, and reference queries per second (QPS) values.
Instance specifications
2 replicas (1 read replica)
| Specification | InstanceClass (for API calls) | CPU cores | I/O threads | Read-only nodes | Bandwidth (MB/s) | Maximum connections (proxy mode) | Reference QPS |
|---|---|---|---|---|---|---|---|
| 1 GB read/write splitting edition | redis.amber.logic.splitrw.small.1db.1rodb.6proxy.multithread | 12 | 4 | 1 | 192 | 60,000 | 480,000 |
| 2 GB read/write splitting edition | redis.amber.logic.splitrw.mid.1db.1rodb.6proxy.multithread | 12 | 4 | 1 | 192 | 60,000 | 480,000 |
| 4 GB read/write splitting edition | redis.amber.logic.splitrw.stand.1db.1rodb.6proxy.multithread | 12 | 4 | 1 | 192 | 60,000 | 480,000 |
| 8 GB read/write splitting edition | redis.amber.logic.splitrw.large.1db.1rodb.6proxy.multithread | 12 | 4 | 1 | 192 | 60,000 | 480,000 |
| 16 GB read/write splitting edition | redis.amber.logic.splitrw.2xlarge.1db.1rodb.6proxy.multithread | 12 | 4 | 1 | 192 | 60,000 | 480,000 |
| 32 GB read/write splitting edition | redis.amber.logic.splitrw.4xlarge.1db.1rodb.6proxy.multithread | 12 | 4 | 1 | 192 | 60,000 | 480,000 |
| 64 GB read/write splitting edition | redis.amber.logic.splitrw.8xlarge.1db.1rodb.6proxy.multithread | 12 | 4 | 1 | 192 | 60,000 | 480,000 |
4 replicas (3 read replicas)
| Specification | InstanceClass (for API calls) | CPU cores | I/O threads | Read-only nodes | Bandwidth (MB/s) | Maximum connections (proxy mode) | Reference QPS |
|---|---|---|---|---|---|---|---|
| 1 GB read/write splitting edition | redis.amber.logic.splitrw.small.1db.3rodb.12proxy.multithread | 24 | 4 | 3 | 384 | 120,000 | 960,000 |
| 2 GB read/write splitting edition | redis.amber.logic.splitrw.mid.1db.3rodb.12proxy.multithread | 24 | 4 | 3 | 384 | 120,000 | 960,000 |
| 4 GB read/write splitting edition | redis.amber.logic.splitrw.stand.1db.3rodb.12proxy.multithread | 24 | 4 | 3 | 384 | 120,000 | 960,000 |
| 8 GB read/write splitting edition | redis.amber.logic.splitrw.large.1db.3rodb.12proxy.multithread | 24 | 4 | 3 | 384 | 120,000 | 960,000 |
| 16 GB read/write splitting edition | redis.amber.logic.splitrw.2xlarge.1db.3rodb.12proxy.multithread | 24 | 4 | 3 | 384 | 120,000 | 960,000 |
| 32 GB read/write splitting edition | redis.amber.logic.splitrw.4xlarge.1db.3rodb.12proxy.multithread | 24 | 4 | 3 | 384 | 120,000 | 960,000 |
| 64 GB read/write splitting edition | redis.amber.logic.splitrw.8xlarge.1db.3rodb.12proxy.multithread | 24 | 4 | 3 | 384 | 120,000 | 960,000 |
6 replicas (5 read replicas)
| Specification | InstanceClass (for API calls) | CPU cores | I/O threads | Read-only nodes | Bandwidth (MB/s) | Maximum connections (proxy mode) | Reference QPS |
|---|---|---|---|---|---|---|---|
| 1 GB read/write splitting edition | redis.amber.logic.splitrw.small.1db.5rodb.18proxy.multithread | 36 | 4 | 5 | 576 | 480,000 | 1,440,000 |
| 2 GB read/write splitting edition | redis.amber.logic.splitrw.mid.1db.5rodb.18proxy.multithread | 36 | 4 | 5 | 576 | 480,000 | 1,440,000 |
| 4 GB read/write splitting edition | redis.amber.logic.splitrw.stand.1db.5rodb.18proxy.multithread | 36 | 4 | 5 | 576 | 480,000 | 1,440,000 |
| 8 GB read/write splitting edition | redis.amber.logic.splitrw.large.1db.5rodb.18proxy.multithread | 36 | 4 | 5 | 576 | 480,000 | 1,440,000 |
| 16 GB read/write splitting edition | redis.amber.logic.splitrw.2xlarge.1db.5rodb.18proxy.multithread | 36 | 4 | 5 | 576 | 480,000 | 1,440,000 |
| 32 GB read/write splitting edition | redis.amber.logic.splitrw.4xlarge.1db.5rodb.18proxy.multithread | 36 | 4 | 5 | 576 | 480,000 | 1,440,000 |
| 64 GB read/write splitting edition | redis.amber.logic.splitrw.8xlarge.1db.5rodb.18proxy.multithread | 36 | 4 | 5 | 576 | 480,000 | 1,440,000 |
Usage notes
CPU cores
The system reserves one CPU core per shard or read replica for background tasks. This applies to both cluster instances and read/write splitting instances.
Bandwidth
Bandwidth figures represent the total for the entire instance — the sum across all shards and nodes.
The maximum total bandwidth for a read/write splitting instance is 2,048 MB/s. Bandwidth does not increase beyond this ceiling regardless of how many nodes the instance has.
Bandwidth applies equally to upstream and downstream traffic. An instance with 96 MB/s bandwidth supports 96 MB/s inbound and 96 MB/s outbound.
Bandwidth limits apply to shard-level throughput and are independent of the network connection type.
To increase bandwidth beyond the default for traffic spikes, see Manually increase the bandwidth of an instance.
For additional details, see the Bandwidth FAQ.
Maximum connections
In proxy mode, the built-in proxy aggregates client connections before forwarding them to data shards. This architecture sets a fixed upper bound on total connections per instance, regardless of node count:
The maximum is 480,000 connections per instance in proxy mode.
For instances created before March 1, 2020, the proxy mode limit is 200,000.
When using PUBSUB, BLOCK, or transaction commands in proxy mode, the proxy creates a dedicated backend connection per client connection — these cannot be aggregated. The effective limit in these cases is the single data shard connection limit of 30,000.