Specifications for memory-optimized Tair (Enterprise Edition) instances in cluster architecture, covering memory per shard, connections, bandwidth, and queries per second (QPS) reference values.
Instance types
Available shard counts: 2, 4, 8, 16, 32, 64, 128, and 256. All instance types allocate 6 CPU cores per shard. One core per shard is reserved for background tasks.
2 shards
12 CPU cores total (6 per shard).
| Instance type name | Instance type code (for API calls) | Memory per shard (GB) | Proxy mode — max connections | Direct connection mode — max connections | Total bandwidth | Bandwidth per shard | QPS reference |
|---|---|---|---|---|---|---|---|
| 2 GB cluster edition (2 shards) | redis.amber.logic.sharding.1g.2db.0rodb.6proxy.multithread | 1 | 360,000 | 60,000 | 1,536 Mbps (192 MB/s) | 768 Mbps (96 MB/s) | 480,000 |
| 4 GB cluster edition (2 shards) | redis.amber.logic.sharding.2g.2db.0rodb.6proxy.multithread | 2 | 360,000 | 60,000 | 1,536 Mbps (192 MB/s) | 768 Mbps (96 MB/s) | 480,000 |
4 shards
24 CPU cores total (6 per shard).
| Instance type name | Instance type code (for API calls) | Memory per shard (GB) | Proxy mode — max connections | Direct connection mode — max connections | Total bandwidth | Bandwidth per shard | QPS reference |
|---|---|---|---|---|---|---|---|
| 8 GB cluster edition (4 shards) | redis.amber.logic.sharding.2g.4db.0rodb.12proxy.multithread | 2 | 500,000 | 120,000 | 3,072 Mbps (384 MB/s) | 768 Mbps (96 MB/s) | 960,000 |
8 shards
48 CPU cores total (6 per shard).
| Instance type name | Instance type code (for API calls) | Memory per shard (GB) | Proxy mode — max connections | Direct connection mode — max connections | Total bandwidth | Bandwidth per shard | QPS reference |
|---|---|---|---|---|---|---|---|
| 16 GB cluster edition (8 shards) | redis.amber.logic.sharding.2g.8db.0rodb.24proxy.multithread | 2 | 500,000 | 240,000 | 6,144 Mbps (768 MB/s) | 768 Mbps (96 MB/s) | 1,920,000 |
| 32 GB cluster edition (8 shards) | redis.amber.logic.sharding.4g.8db.0rodb.24proxy.multithread | 4 | 500,000 | 240,000 | 6,144 Mbps (768 MB/s) | 768 Mbps (96 MB/s) | 1,920,000 |
| 64 GB cluster edition (8 shards) | redis.amber.logic.sharding.8g.8db.0rodb.24proxy.multithread | 8 | 500,000 | 240,000 | 6,144 Mbps (768 MB/s) | 768 Mbps (96 MB/s) | 1,920,000 |
| 128 GB cluster edition (8 shards) | redis.amber.logic.sharding.16g.8db.0rodb.24proxy.multithread | 16 | 500,000 | 240,000 | 6,144 Mbps (768 MB/s) | 768 Mbps (96 MB/s) | 1,920,000 |
16 shards
96 CPU cores total (6 per shard).
| Instance type name | Instance type code (for API calls) | Memory per shard (GB) | Proxy mode — max connections | Direct connection mode — max connections | Total bandwidth | Bandwidth per shard | QPS reference |
|---|---|---|---|---|---|---|---|
| 32 GB cluster edition (16 shards) | redis.amber.logic.sharding.2g.16db.0rodb.48proxy.multithread | 2 | 500,000 | 480,000 | 12,288 Mbps (1,536 MB/s) | 768 Mbps (96 MB/s) | 3,840,000 |
| 128 GB cluster edition (16 shards) | redis.amber.logic.sharding.8g.16db.0rodb.48proxy.multithread | 8 | 500,000 | 480,000 | 12,288 Mbps (1,536 MB/s) | 768 Mbps (96 MB/s) | 3,840,000 |
| 256 GB cluster edition (16 shards) | redis.amber.logic.sharding.16g.16db.0rodb.48proxy.multithread | 16 | 500,000 | 480,000 | 12,288 Mbps (1,536 MB/s) | 768 Mbps (96 MB/s) | 3,840,000 |
32 shards
192 CPU cores total (6 per shard).
| Instance type name | Instance type code (for API calls) | Memory per shard (GB) | Proxy mode — max connections | Direct connection mode — max connections | Total bandwidth | Bandwidth per shard | QPS reference |
|---|---|---|---|---|---|---|---|
| 64 GB cluster edition (32 shards) | redis.amber.logic.sharding.2g.32db.0rodb.96proxy.multithread | 2 | 500,000 | 960,000 | 16,384 Mbps (2,048 MB/s) | 768 Mbps (96 MB/s) | 7,680,000 |
| 512 GB cluster edition (32 shards) | redis.amber.logic.sharding.16g.32db.0rodb.96proxy.multithread | 16 | 500,000 | 960,000 | 16,384 Mbps (2,048 MB/s) | 768 Mbps (96 MB/s) | 7,680,000 |
64 shards
384 CPU cores total (6 per shard).
| Instance type name | Instance type code (for API calls) | Memory per shard (GB) | Proxy mode — max connections | Direct connection mode — max connections | Total bandwidth | Bandwidth per shard | QPS reference |
|---|---|---|---|---|---|---|---|
| 128 GB cluster edition (64 shards) | redis.amber.logic.sharding.2g.64db.0rodb.192proxy.multithread | 2 | 500,000 | 1,920,000 | 16,384 Mbps (2,048 MB/s) | 768 Mbps (96 MB/s) | 15,360,000 |
| 256 GB cluster edition (64 shards) | redis.amber.logic.sharding.4g.64db.0rodb.192proxy.multithread | 4 | 500,000 | 1,920,000 | 16,384 Mbps (2,048 MB/s) | 768 Mbps (96 MB/s) | 15,360,000 |
| 1024 GB cluster edition (64 shards) | redis.amber.logic.sharding.16g.64db.0rodb.192proxy.multithread | 16 | 500,000 | 1,920,000 | 16,384 Mbps (2,048 MB/s) | 768 Mbps (96 MB/s) | 15,360,000 |
128 shards
768 CPU cores total (6 per shard).
| Instance type name | Instance type code (for API calls) | Memory per shard (GB) | Proxy mode — max connections | Direct connection mode — max connections | Total bandwidth | Bandwidth per shard | QPS reference |
|---|---|---|---|---|---|---|---|
| 2048 GB cluster edition (128 shards) | redis.amber.logic.sharding.16g.128db.0rodb.384proxy.multithread | 16 | 500,000 | 3,840,000 | 16,384 Mbps (2,048 MB/s) | 768 Mbps (96 MB/s) | 30,720,000 |
256 shards
1,536 CPU cores total (6 per shard).
| Instance type name | Instance type code (for API calls) | Memory per shard (GB) | Proxy mode — max connections | Direct connection mode — max connections | Total bandwidth | Bandwidth per shard | QPS reference |
|---|---|---|---|---|---|---|---|
| 4096 GB cluster edition (256 shards) | redis.amber.logic.sharding.16g.256db.0rodb.768proxy.multithread | 16 | 500,000 | 7,680,000 | 16,384 Mbps (2,048 MB/s) | 768 Mbps (96 MB/s) | 61,440,000 |
Usage notes
Bandwidth
Total bandwidth equals the sum of all shard bandwidths. Each shard has a maximum of 768 Mbps (96 MB/s).
When using the default database proxy endpoint, total bandwidth is capped at 16,384 Mbps (2,048 MB/s) regardless of shard count.
Bandwidth applies symmetrically to both upstream and downstream traffic.
Bandwidth limits apply to shard nodes and are independent of network connection type.
To exceed the proxy endpoint cap, enable direct connection mode. In direct connection mode, total bandwidth equals 768 Mbps x number of shards. For example, a 128-shard instance where each shard has more than 1 GB of memory achieves 12,288 MB/s total bandwidth in direct connection mode. See Enable direct connection mode (cluster instances only).
To handle burst or planned traffic peaks, adjust instance bandwidth as needed. See Manually increase the bandwidth of an instance.
Maximum connections
Proxy mode Capped at 500,000 per instance. Adding more shards or nodes does not increase this limit.
Direct connection mode 30,000 connections per shard. Total = number of shards x 30,000.
For instances created before March 1, 2020, the proxy mode cap is 200,000.
When using PUBSUB, blocking commands, or transaction commands in proxy mode, the proxy creates dedicated backend connections per client — connections cannot be aggregated. The effective limit drops to the per-shard maximum, which is the same threshold as direct connection mode.
CPU cores
The system reserves one CPU core per shard (or read replica) to handle background tasks in cluster and read/write splitting instances.
FAQ
Why does the actual bandwidth of my instance differ from the values listed here?
Bandwidth for specific instance types may be updated after a service upgrade. If your instance bandwidth no longer matches the table, change the instance configuration by selecting the same instance type — this refreshes the bandwidth at no additional cost. See Change instance configurations.
Why does write traffic fail even though the instance still has available memory?
Cluster instances use a hash algorithm to distribute data across shards. If the instance contains big keys, data distributes unevenly and individual shards can become full while overall instance memory remains available. When a shard is exhausted, write requests targeting that shard fail.
Use Performance monitoring to check memory usage per shard and identify hot shards. See View the performance metrics of nodes in a Tair cluster instance.
Does the number after "proxy" in an instance type code (for example, `8proxy`) indicate the number of proxy nodes?
No. The proxy value in the instance type code does not represent a node count. Proxy capacity is managed automatically to meet the bandwidth and compute requirements listed in the spec table.