Performance monster on arm64: installation and performance test of API gateway Apache APIs IX on AWS graviton3

background

AWS stay 2022 year 5 At the end of the month, the latest version based on ARM Architecturally AWS Graviton Family processor ——AWS Graviton3. According to the AWS Official data show , And Graviton2 Processor comparison , Based on leading DDR5 Memory technology ,Graviton3 The processor can provide up to 25% Performance improvement of 、 the height is 2 Times the floating-point performance and 50% Memory access speed ; The performance is similar to that of the same kind EC2 With the same instance ,Graviton3 Can also reduce 60% Energy .

So what about the actual data ？ Let's take CPU intensive API For example, gateway , Let's see AWS Graviton3 How do you behave . Here we use Apache APISIX stay AWS Graviton2（C6g） and AWS Graviton3（C7g） Performance comparison test under two server environments .

Apache APISIX It's a cloud native 、 High performance 、 Extensible API gateway . be based on NGNIX+LuaJIT and etcd To achieve , And traditional API Gateway compared to ,APISIX It has the characteristics of dynamic routing and plug-in hot loading , It is especially suitable for the cloud native architecture API management .

Get ready ： Installation and deployment

Before testing , It is necessary to prepare one to carry ARM64 Chip servers , Here we choose Amazon EC2 C7g（ Now only this model is equipped with AWS Graviton3）, The operating system chose Ubuntu 20.04.

At the same time to install Docker.

sudo apt-get update && sudo apt-get install docker.io

at present ,APISIX The latest version of ARM64 Mirror image , have access to Docker One click deployment . Refer to the following for the specific process ：

1. start-up etcd

sudo docker run -d \--name etcd -p 2379:2379 -e ETCD_UNSUPPORTED_ARCH=arm64 \-e ETCD_LISTEN_CLIENT_URLS=http://0.0.0.0:2379 \-e ETCD_ADVERTISE_CLIENT_URLS=http://0.0.0.0:2379 \rancher/coreos-etcd:v3.4.16-arm64

2. start-up APISIX

sudo docker run --net=host -d apache/apisix:2.14.1-alpine

3. Registered routing

curl "http://127.0.0.1:9080/apisix/admin/routes/1" \-H "X-API-KEY: edd1c9f034335f136f87ad84b625c8f1" -X PUT -d '{  "uri": "/anything/*",  "upstream": {    "type": "roundrobin",    "nodes": {      "httpbin.org:80": 1    }  }}'

4. Access test

curl -i http://127.0.0.1:9080/anything/das

HTTP/1.1 200 OK.....

AWS Graviton2 and AWS Graviton3 Performance comparison of

According to the above operation , be based on Official script It's done APISIX stay AWS Graviton3 Installation and compatibility testing on the processor . Let's take a look Apache APISIX stay AWS Graviton2（C6g） and AWS Graviton3（C7g） Performance on .

For testing purposes , In this example APISIX Only one... Has been opened Worker, The following performance test data are in the single core CPU Running on .

Scene one ： Single upstream

A single upstream... Is used in this scenario （ Does not contain any plug-ins ）, Main tests APISIX Performance in pure proxy back to source mode . Testing in a local environment ：

# apisix: 1 worker + 1 upstream + no plugin#  Registered routing curl http://127.0.0.1:9080/apisix/admin/routes/1 \-H 'X-API-KEY: edd1c9f034335f136f87ad84b625c8f1' -X PUT -d '{    "uri": "/hello",    "plugins": {    },    "upstream": {        "type": "roundrobin",        "nodes": {            "127.0.0.1:1980":1        }    }}'

Scene two ： Single upstream + Multiple plug-ins

The other scenario uses single upstream and multiple plug-ins , Two plug-ins are used here . Main tests APISIX In the open limit-count and prometheus Performance performance of two cores when consuming performance plug-ins .

# apisix: 1 worker + 1 upstream + 2 plugins (limit-count + prometheus)#  Registered routing curl http://127.0.0.1:9080/apisix/admin/routes/1 \-H 'X-API-KEY: edd1c9f034335f136f87ad84b625c8f1' -X PUT -d '{    "uri": "/hello",    "plugins": {        "limit-count": {            "count": 2000000000000,            "time_window": 60,            "rejected_code": 503,            "key": "remote_addr"        },        "prometheus": {}    },    "upstream": {        "type": "roundrobin",        "nodes": {            "127.0.0.1:1980":1        }    }}'

Data comparison

In the above two scenarios , The related tests and comparisons are carried out from the two aspects of request processing and delay time . give the result as follows ：

1. QPS contrast

2. Latency contrast

From the data above, you can see , stay API The gateway is like this CPU In a dense computing scenario ,AWS Graviton3 Than AWS Graviton2 The performance of 76%, The delay is also reduced 38%. This figure is better than the one mentioned at the beginning AWS Official figures （25% Performance improvement ） And excellent .

summary

This paper mainly uses Apache APISIX the AWS Graviton3 And AWS Graviton2 Performance comparison of , You can see in the API gateway CPU In a dense computing scenario ,AWS Graviton3 It shows the properties of the performance monster . Of course , I also recommend you to practice more , We look forward to more test data of computing intensive projects in the future .