Performance whitepaper of SSD-based instances - Tair - Alibaba Cloud Documentation Center

This topic describes the test environment, test tool, test method, and test results of a performance test on Tair SSD-based instances.

Test environment

Item	Description
Region and zone	All tests are performed in Hangzhou Zone I.
Instance architecture	The standard master-replica architecture is used. For more information, see Standard architecture.
Virtual machine on which a stress testing tool is deployed	An Elastic Compute Service (ECS) instance of the ecs.g6e.13xlarge instance type is used. For more information, see Overview of instance families.
SSD-based instance type	tair.localssd.c1m4.2xlarge tair.localssd.c1m4.4xlarge tair.localssd.c1m4.8xlarge For more information about the instance types, see ESSD/SSD-based instances.

The performance test focuses on evaluating the performance of SSD-based instances in the following scenarios:

Scenarios in which the memory size is larger than the data volume: Most data can be accessed from the memory. In these scenarios, the ratio of the memory size to the data volume is approximately 7:1.
Scenarios in which the data volume is larger than the memory size: Only part of the data is cached in the memory. Most access requests require read and write operations on disks. In these scenarios, the ratio of the memory size to the data volume is approximately 1:4.

Test tool

You can use open source YCSB to perform stress tests. YCSB is a Java tool that can be used to benchmark the performance of multiple types of databases. For more information about how to install and use YCSB, see YCSB.

In the test, the YCSB source code is modified to allow you to set the recordcount parameter to a LONG-typed value and test the string-related commands of Redis. To obtain the modified source code, download the YCSB source code package.

Test commands

The following script is used to test performance in scenarios where the data volume is larger than the memory size:

#! /bin/bash

ip=192.168.0.23
port=3100
timeout=30000
command_group=string
recordcount=640000000
run_operationcount=20000000
fieldcount=1
fieldlength=100
threads=32
load_sleep_time=600
run_sleep_time=60

echo "##################################### $command_group  ############################################"
#Load
./bin/ycsb load redis -s -P workloads/workloada -p "redis.host=${ip}" -p  "redis.port=${port}" -p "recordcount=${recordcount}" -p "operationcount=${recordcount}" -p "redis.timeout=${timeout}" -p "redis.command_group=${command_group}" -p "fieldcount=${fieldcount}" -p "fieldlength=${fieldlength}" -threads ${threads}
sleep ${load_sleep_time}

#Uniform-Read
./bin/ycsb run redis -s -P workloads/workloadc  -p "redis.host=${ip}" -p  "redis.port=${port}" -p "recordcount=${recordcount}" -p "operationcount=${run_operationcount}" -p "redis.timeout=${timeout}" -p "redis.command_group=${command_group}" -p "fieldcount=${fieldcount}" -p "fieldlength=${fieldlength}" -p "requestdistribution=uniform" -threads ${threads}
sleep ${run_sleep_time}

#Zipfian-Read
./bin/ycsb run redis -s -P workloads/workloadc  -p "redis.host=${ip}" -p  "redis.port=${port}" -p "recordcount=${recordcount}" -p "operationcount=${run_operationcount}" -p "redis.timeout=${timeout}" -p "redis.command_group=${command_group}" -p "fieldcount=${fieldcount}" -p "fieldlength=${fieldlength}" -p "requestdistribution=zipfian" -threads ${threads}
sleep ${run_sleep_time}

#Uniform-50%Read-50%Update
./bin/ycsb run redis -s -P workloads/workloada  -p "redis.host=${ip}" -p  "redis.port=${port}" -p "recordcount=${recordcount}" -p "operationcount=${run_operationcount}" -p "redis.timeout=${timeout}" -p "redis.command_group=${command_group}" -p "fieldcount=${fieldcount}" -p "fieldlength=${fieldlength}" -p "requestdistribution=uniform" -threads ${threads}

Table 1. Parameters

Parameter	Description
ip	The IP address of the Tair instance.
port	The service port of the Tair instance.
timeout	The timeout period of the test command. Unit: ms.
command_group	The type of data to be tested. Set this parameter to String.
recordcount	The amount of data prepared during the data loading phase.
run_operationcount	The amount of data involved in the operation. When you set this parameter, take note of the following rules: In scenarios in which the memory size is larger than the data volume, set this parameter to the value of the recordcount parameter. In scenarios in which the data volume is larger than the memory size, set this parameter to the value of the recordcount parameter divided by 32.
fieldcount	The number of fields. Set this parameter to 1.
fieldlength	The length of each field. Set this parameter to 100.
threads	The number of YCSB threads. Set this parameter based on the instance type.

Test results

Test metric	Description
QPS	The number of read and write operations that are processed per second.
Average Latency	The average latency of read or write operations. Unit: µs.
99th Percentile Latency	The highest latency for the fastest 99% of operations. Unit: µs. For example, if this parameter is set to 500 µs, 99% of the operations are processed within 500 µs.

Scenarios in which the memory size is larger than the data volume

Instance type	YCSB configuration	Workload	QPS	Average latency (µs)	99th percentile latency (µs)
tair.localssd.c1m4.2xlarge	recordcount=40000000 run_operationcount=40000000 threads=64	Load	59,830	1,066	2,761
		Uniform-Read	158,221	389	891
		Zipfian-Read	164,233	379	873
		Uniform-50%Read-50%Update	78,099	Read: 651	Read: 2,012
		Uniform-50%Read-50%Update	78,099	Update: 974	Update: 2,731
tair.localssd.c1m4.4xlarge	recordcount=80000000 run_operationcount=80000000 threads=128	Load	91,991	1,388	3,077
		Uniform-Read	302,940	414	921
		Zipfian-Read	305,639	410	899
		Uniform-50%Read-50%Update	124,929	Read: 798	Read: 2,231
		Uniform-50%Read-50%Update	124,929	Update: 1,234	Update: 3,013
tair.localssd.c1m4.8xlarge	recordcount=160000000 run_operationcount=160000000 threads=256	Load	132,865	1,924	3,323
		Uniform-Read	489,287	513	1,313
		Zipfian-Read	501,847	499	1,272
		Uniform-50%Read-50%Update	187,390	Read: 1,069	Read: 2,749
		Uniform-50%Read-50%Update	187,390	Update: 1,644	Update: 3,613

Scenarios in which the data volume is larger than the memory size

Instance type	YCSB configuration	Workload	QPS	Average latency (µs)	99th percentile latency (µs)
tair.localssd.c1m4.2xlarge	recordcount=1280000000 run_operationcount=1280000000 threads=64	Load	50,396	1,258	4,463
		Uniform-Read	74,611	842	1,745
		Zipfian-Read	106,366	588	1,406
		Uniform-50%Read-50%Update	47,833	Read: 1,232	Read: 4,049
		Uniform-50%Read-50%Update	47,833	Write: 1,402	Write: 4,583
tair.localssd.c1m4.4xlarge	recordcount=2560000000 run_operationcount=2560000000 threads=128	Load	81,097	1,573	4,119
		Uniform-Read	118,141	1,071	3,085
		Zipfian-Read	194,704	634	1,595
		Uniform-50%Read-50%Update	75,625	Read: 1,562	Read: 4,999
		Uniform-50%Read-50%Update	75,625	Update: 1,795	Update: 5,419
tair.localssd.c1m4.8xlarge	recordcount=5120000000 run_operationcount=5120000000 threads=256	Load	115,660	2,210	5,235
		Uniform-Read	202,365	1,252	3,985
		Zipfian-Read	309,019	804	2,551
		Uniform-50%Read-50%Update	122,318	Read: 1,861	Read: 5,603
		Uniform-50%Read-50%Update	122,318	Update: 2,307	Update: 6,415

References

ESSD/SSD-based instances