Performance white paper of persistent memory-optimized instances - Tair (Redis® OSS-Compatible)

This topic describes the tests on Tair (Enterprise Edition) persistent memory-optimized instances. The test environment, tool, methods, and results are introduced.

Overview

Tair persistent memory-optimized instances adopt persistent memory to provide large-capacity in-memory databases that are compatible with open source Redis. Persistent memory-optimized instances do not use disks to implement data persistence and cost 30% less than ApsaraDB for Redis Community Edition instances while delivering almost identical throughput and latency and maintaining the persistence of each operation. This helps improve the reliability of business data. For more information, see Persistent memory-optimized instances.

Test environment

Item	Description
Region and zone	All tests are performed in Zhangjiakou Zone A.
Instance architecture	The standard master-replica architecture is used. For more information, see Standard master-replica instances.
Virtual machine on which a stress testing tool is deployed	An Elastic Compute Service (ECS) instance of the ecs.g6e.8xlarge type is used. For more information, see Overview of instance families.
Instance type	In the test, the instance type is tair.scm.standard.32m.128d.

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.

Open source YCSB tests only hashes. To use YCSB to test strings, you must use the RedisClient.java file to replace the original file that has the same name in the YCSB/redis/src/main/java/site/ycsb/db directory after you download YCSB.

Workloads

The total amount of data involved in the test is 8 GB. The data distribution method is Zipfian. The performance of persistent memory-optimized instances is tested in the following scenarios:

Data loading: 100% write operations.
Workload C: 100% read operations.
Workload A: 50% update operations and 50% read operations.

For more information about workloads, see Core Workloads.

Note

The complex commands for sets and zsets are not suitable for this test due to their applicable scenarios. You can customize a test for sets and zsets based on your business scenario.

Test commands

# Load data.
workload=a
./bin/ycsb load redis -s -P workloads/workload${workload} -p "redis.host=${server_ip}" -p  "redis.port=${port}" -p "redis.password=${password}" -p "recordcount=${recordcount}" -p "operationcount=${operationcount}" -p "redis.timeout=30000" -p "redis.command_group=${command_group}" -p "fieldcount=${fieldcount}" -p "fieldlength=${fieldlength}" -threads ${threads}  -p "redis.password=***:*****"

# Run Workload C.
workload=c
./bin/ycsb run redis -s -P workloads/workload${workload} -p "redis.host=${server_ip}" -p  "redis.port=${port}" -p "redis.password=${password}" -p "recordcount=${recordcount}" -p "operationcount=${operationcount}" -p "redis.timeout=30000" -p "redis.command_group=${command_group}" -p "fieldcount=${fieldcount}" -p "fieldlength=${fieldlength}" -threads ${threads} -p "redis.password=***:*****"

# Run Workload A.
workload=a
./bin/ycsb run redis -s -P workloads/workload${workload} -p "redis.host=${server_ip}" -p  "redis.port=${port}" -p "redis.password=${password}" -p "recordcount=${recordcount}" -p "operationcount=${operationcount}" -p "redis.timeout=30000" -p "redis.command_group=${command_group}" -p "fieldcount=${fieldcount}" -p "fieldlength=${fieldlength}" -threads ${threads} -p "redis.password=***:*****"

Table 1. Parameters

Parameter	Description
server_ip	The IP address of the instance.
port	The service port of the instance.
password	The password format varies based on the selected account: If you use the default account, enter the password. For example, if the username of the default account is `r-bp1zxszhcgatnx**` and the password is `Password21`, the command used to verify the password is `AUTH Password21`. If you use a custom account, enter the password in the `user:password` format. For example, if the username of the custom account is `testaccount` and the password is `Rp829dlwa`, the command used to verify the password is `AUTH testaccount:Rp829dlwa`. Note** If you forget your password, you can reset it. For more information, see Change or reset the password.
recordcount	The amount of prepared data.
operationcount	The amount of data involved in the operation.
command_group	The type of data that you want to test. In this test, the following types of data are tested in sequence: String Hash List Set Zset
fieldcount	The number of fields or elements. To test strings, set this parameter to 1. To test other types of data, set this parameter to 10.
fieldlength	The value length. Set this parameter based on the test requirements.
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.
INSERT Average Latency	The average latency of write operations. Unit: µs.
INSERT 99th Percentile Latency	The highest latency for the fastest 99% of write operations. Unit: µs. For example, if this parameter is set to 500 µs, 99% of the operations are processed within 500 µs.
READ AverageLatency	The average latency of read operations. Unit: µs.
READ 99thPercentileLatency	The highest latency for the fastest 99% of read operations. Unit: µs.
UPDATE AverageLatency	The average latency of update operations. Unit: µs.
UPDATE 99thPercentileLatency	The highest latency for the fastest 99% of update operations. Unit: µs.

Test results for data loading

Table 2. Strings

Value length (bytes)	QPS	INSERT average latency (µs)	INSERT 99th percentile latency (µs)
128	134,478	473	687
256	126,139	504	828
1,024	99,775	638	1,051
2,048	77,130	826	1,157
4,096	60,646	1,050	1,534

Table 3. Hashes

Number of fields in a key	Value length (bytes)	QPS	INSERT average latency (µs)	INSERT 99th percentile latency (µs)
10	128	47,353	1,348	1,885
	256	46,716	1,366	2,181
	1,024	27,759	2,297	2,873
	2,048	16,605	3,833	4,923

Table 4. Lists

Number of elements in a key	Value length (bytes)	QPS	INSERT average latency (µs)	INSERT 99th percentile latency (µs)
10	128	64,950	979	1,310
	256	47,157	1,348	1,752
	1,024	26,719	2,386	3,457
	2,048	16,714	3,811	4,751
	4,096	10,129	6,279	7,891

Table 5. Sets

Number of members in a key	Value length (bytes)	QPS	INSERT average latency (µs)	INSERT 99th percentile latency (µs)
10	128	63,670	1,001	1,514
	256	44,707	1,427	2,129
	1,024	25,375	2,513	3,239
	2,048	14,318	4,451	5,619
	4,096	8,378	7,608	9,095

Table 6. Sorted sets

Number of elements in a key	Value length (bytes)	QPS	INSERT average latency (µs)	INSERT 99th percentile latency (µs)
10	128	40,292	1,585	2,469
	256	34,168	1,869	2,569
	1,024	21,347	2,989	3,905
	2,048	12,868	4,956	6,255
	4,096	7,864	8,101	9,599

Test results for Workload C

Table 7. Strings

Value length (bytes)	QPS	READ average latency (µs)	READ 99th percentile latency (µs)
128	170,699	362	546
256	163,829	380	565
1,024	161,491	386	569
2,048	130,189	487	729
4,096	115,433	548	808

Table 8. Hashes

Number of fields in a key	Value length (bytes)	QPS	READ average latency (µs)	READ 99th percentile latency (µs)
10	128	96,111	662	874
	256	86,892	733	915
	1,024	61,608	1,030	1,293
	2,048	37,334	1,696	2,331
	4,096	25,943	2,429	3,319

Table 9. Lists

Number of elements in a key	Value length (bytes)	QPS	READ average latency (µs)	READ 99th percentile latency (µs)
10	128	105,296	604	889
	256	97,047	655	890
	1,024	66,384	955	1,192
	2,048	35,796	1,769	2,461
	4,096	26,314	2,392	3,271

Table 10. Sets

Number of members in a key	Value length (bytes)	QPS	READ average latency (µs)	READ 99th percentile latency (µs)
10	128	97,825	651	896
	256	80,954	787	970
	1024	59,924	1060	1313
	2048	33,356	1900	2637
	4096	23,605	2677	3723

Table 11. Sorted sets

Number of elements in a key	Value length (bytes)	QPS	READ average latency (µs)	READ 99th percentile latency (µs)
10	128	58,380	1,093	1,341
	256	56,287	1,133	1,390
	1,024	47,468	1,338	1,688
	2,048	30,073	2,096	2,783
	4,096	21,850	2,880	3,765

Test results for Workload A

Table 12. Strings

Value length (bytes)	QPS	READ average latency (µs)	READ 99th percentile latency (µs)	UPDATE average latency (µs)	UPDATE 99th percentile latency (µs)
128	141,120	451	616	450	618
256	137,551	463	617	461	618
1,024	124,165	516	724	508	725
2,048	92,652	695	881	678	871
4,096	78,994	819	1,042	791	1,024

Table 13. Hashes

Number of fields in a key	Value length (bytes)	QPS	READ average latency (µs)	READ 99th percentile latency (µs)	UPDATE average latency (µs)	UPDATE 99th percentile latency (µs)
10	128	99,495	646	831	633	820
	256	88,235	731	985	712	966
	1,024	72,013	892	1,159	863	2,049
	2,048	45,790	1,379	1,898	1,354	2,821
	4,096	32,912	1,891	2,931	1,915	7,887

Table 14. Lists

Number of elements in a key	Value length (bytes)	QPS	READ average latency (µs)	READ 99th percentile latency (µs)	UPDATE average latency (µs)	UPDATE 99th percentile latency (µs)
10	128	71,696	591	775	1,185	1,383
	256	66,294	638	800	1,281	1,456
	1,024	53,402	791	1,006	1,581	1,865
	2,048	36,519	1,221	1,581	2,232	2,831
	4,096	28,390	1,618	2,113	2,803	3,777

Table 15. Sets

Number of members in a key	Value length (bytes)	QPS	READ average latency (µs)	READ 99th percentile latency (µs)	UPDATE average latency (µs)	UPDATE 99th percentile latency (µs)
10	128	66,346	640	792	1,282	1,445
	256	60,010	707	993	1,415	1,957
	1,024	45,359	933	1,128	1,858	2,073
	2,048	29,027	1,529	2,021	2,820	4,507
	4,096	21,440	2,144	2,773	3,726	5,095

Table 16. Sorted sets

Number of elements in a key	Value length (bytes)	QPS	READ average latency (µs)	READ 99th percentile latency (µs)	UPDATE average latency (µs)	UPDATE 99th percentile latency (µs)
10	128	49,695	861	1,050	1,707	1,912
	256	48,036	891	1,084	1,763	1,970
	1,024	39,795	1,081	1,386	2,107	2,563
	2,048	28,415	1,597	1,981	2,855	3,589
	4,096	21,317	2,247	2,821	3,665	4,787