Cloud-native gateway capacity and performance

更新时间:
复制 MD 格式

This topic provides capacity thresholds and Queries Per Second (QPS) performance data for different specifications of standard cloud-native gateway instances to help you with your selection. Use this reference to select the right gateway specification for your workload. Each specification defines per-node capacity thresholds and queries-per-second (QPS) benchmarks. In addition, for Serverless instances of the cloud-native gateway, you do not need to perform a detailed capacity assessment. The system automatically performs elastic scaling and billing based on your business traffic, and the scaling process includes a protection threshold to prevent uncontrolled costs.

Select a specification

Match two dimensions to your workload: capacity thresholds (connections, bandwidth, resource utilization) and QPS throughput (request processing rate).

  1. Check capacity thresholds to find the specification that handles your connection count, bandwidth, and resource usage.

  2. Check QPS benchmarks to verify throughput for your traffic pattern (connection type, HTTPS, GZIP, response size).

  3. If one dimension fits but the other does not, choose the next higher specification.

For a worked example, see Sizing example.

Capacity thresholds

The following tables list per-node capacity thresholds for each gateway specification across three tiers:

TierMeaningAction
Security thresholdThe gateway handles current traffic and can absorb a 2x traffic spike without degradation.No action required. Normal operation.
Warning thresholdLatency may increase. Traffic surges risk dropped connections and timeouts.Monitor closely. Consider adding nodes or upgrading the specification.
Overload thresholdThe gateway rejects new connections to protect itself from failure.Add nodes or upgrade the specification to reduce load.
Important

Deploy at least two nodes per gateway. A single-node deployment may not meet your service-level agreement (SLA) objectives. All thresholds below apply to each individual node.

Client connections

Threshold2 cores, 4 GiB4 cores, 8 GiB8 cores, 16 GiB16 cores, 32 GiB
Security12,00024,00048,00096,000
Warning24,00048,00096,000192,000
Overload40,00080,000160,000320,000

New HTTPS connections per second

Threshold2 cores, 4 GiB4 cores, 8 GiB8 cores, 16 GiB16 cores, 32 GiB
Security4008001,6003,200
Warning8001,6003,2006,400
Overload----

Network bandwidth (Gbit/s)

Threshold2 cores, 4 GiB4 cores, 8 GiB8 cores, 16 GiB16 cores, 32 GiB
Security1248
Warning1248
Overload----

CPU utilization

ThresholdAll specifications
Security30%
Warning60%
Overload90%

Memory usage

ThresholdAll specifications
Security75%
Warning75%
Overload90%

QPS benchmarks

QPS throughput varies based on four factors:

FactorImpact
Connection typePersistent connections deliver higher QPS than short-lived connections by skipping repeated TCP and TLS handshakes.
HTTPSTLS handshakes on new connections are CPU-intensive, reducing QPS significantly for short-lived connections. For persistent connections, the handshake occurs only once, so the impact is smaller.
GZIP compressionCompressing responses adds CPU overhead, which reduces QPS.
Response sizeLarger responses consume more bandwidth and processing time per request.

The following values are pessimistic estimates (worst-case) measured at CPU utilization below 30%.

Note

For workloads with a high rate of simultaneous new HTTPS connections, use the short-lived connection rows to estimate capacity.

Short-lived connections, 1 KB response

HTTPSGZIP2c4g (3 nodes)2c4g (5 nodes)4c8g (3 nodes)4c8g (5 nodes)8c16g (3 nodes)8c16g (5 nodes)16c32g (3 nodes)16c32g (5 nodes)
NoNo5,2008,70010,50017,50021,00035,00042,00070,000
YesNo1,6002,7003,2005,5006,50011,00013,00022,000

Persistent connections, 1 KB response

HTTPSGZIP2c4g (3 nodes)2c4g (5 nodes)4c8g (3 nodes)4c8g (5 nodes)8c16g (3 nodes)8c16g (5 nodes)16c32g (3 nodes)16c32g (5 nodes)
NoNo6,50010,80013,00021,70026,00043,50052,00087,000
YesNo6,00010,00012,00020,00024,00040,00048,00080,000
YesYes5,2008,70010,50017,50021,00035,00042,00070,000

Persistent connections, 10 KB response

HTTPSGZIP2c4g (3 nodes)2c4g (5 nodes)4c8g (3 nodes)4c8g (5 nodes)8c16g (3 nodes)8c16g (5 nodes)16c32g (3 nodes)16c32g (5 nodes)
NoNo5,6009,30011,20018,70022,50037,50045,00075,000
YesNo5,3009,00010,70018,00021,50036,00043,00072,000
YesYes3,1005,2006,20010,50012,50021,00025,00042,000

Sizing example

Suppose your workload has the following characteristics:

  • 30,000 concurrent client connections

  • Persistent HTTPS connections with GZIP compression

  • Average response size: 1 KB

  • Target QPS: 15,000

Step 1: Check capacity thresholds.

A 4-core, 8 GiB node supports 24,000 client connections at the security threshold. At 30,000 connections, you exceed the security threshold but stay below the warning threshold (48,000). For production workloads, choose the 8-core, 16 GiB specification. This keeps 30,000 connections well within the security threshold (48,000).

Step 2: Check QPS benchmarks.

For persistent HTTPS connections with GZIP and 1 KB responses, the 8-core, 16 GiB specification delivers:

  • 3 nodes: 21,000 QPS

  • 5 nodes: 35,000 QPS

A 3-node cluster at 21,000 QPS exceeds the 15,000 QPS target.

Result: Deploy an 8-core, 16 GiB gateway with 3 nodes.