Whitepaper performa instans berbasis SSD - Tair (Redis® OSS-Compatible)

Dokumen ini mencakup lingkungan pengujian, tooling, metodologi, dan hasil benchmark performa instans Tair (Edisi Perusahaan) berbasis SSD.

Lingkungan pengujian

Item	Description
Region and zone	Hangzhou Zone J
Instance architecture	Standard master-replica architecture. Untuk detailnya, lihat Standard architecture.
Stress testing host	Instans Elastic Compute Service (ECS) tipe `ecs.c9i.8xlarge`. Untuk detailnya, lihat Overview of instance families.
Instance types tested	`tair.localssd.c1m8.xlarge`, `tair.localssd.c1m8.2xlarge`, `tair.localssd.c1m8.with.proxy.xlarge (3 shards)`, `tair.localssd.c1m8.with.proxy.2xlarge (3 shards)`

Benchmark ini mencakup dua skenario yang mewakili ujung berlawanan dari spektrum pemanfaatan memori:

Memory larger than data volume: Sebagian besar data muat dalam memori dengan rasio memori terhadap data sekitar 7:1, merepresentasikan pola akses yang didominasi memori.
Data volume larger than memory: Hanya sebagian data yang di-cache di memori, sehingga sebagian besar permintaan baca/tulis memerlukan I/O disk dengan rasio memori terhadap data sekitar 1:4, merepresentasikan pola akses yang didominasi SSD.

Test tool

Benchmark ini menggunakan YCSB, sebuah framework benchmarking database berbasis Java open-source.

Kode sumber YCSB telah dimodifikasi untuk mendukung parameter recordcount bertipe LONG serta menguji perintah string Redis. Untuk kode sumber yang telah dimodifikasi, lihat YCSB source code.

Test script

Skrip berikut menjalankan pengujian performa untuk skenario data volume larger than memory:

#!/bin/bash

ip=192.168.0.23
port=3100
timeout=30000
command_group=string
recordcount=640000000
run_operationcount=640000000
fieldcount=1
fieldlength=100
threads=64
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}

Parameters

Parameter	Description
`ip`	Alamat IP instans Tair
`port`	Port layanan instans Tair
`timeout`	Waktu tunggu perintah. Satuan: ms
`command_group`	Tipe data yang diuji. Diatur ke `string`
`recordcount`	Jumlah catatan yang dimuat selama fase pemuatan data
`run_operationcount`	Jumlah operasi selama fase eksekusi. Pada kedua skenario, atur nilai ini sama dengan `recordcount`.
`fieldcount`	Jumlah bidang per catatan. Diatur ke `1`
`fieldlength`	Panjang setiap bidang dalam byte. Diatur ke `100`
`threads`	Jumlah thread klien YCSB. Berbeda-beda tergantung tipe instans.

Test metrics

Metric	Description
QPS	Jumlah operasi baca dan tulis yang diproses per detik
Average latency	Latensi rata-rata operasi baca atau tulis. Satuan: µs
99th percentile latency	Latensi tertinggi untuk 99% operasi tercepat. Satuan: µs. Misalnya, nilai 500 µs berarti 99% operasi selesai dalam waktu 500 µs.

Test results

Memory size larger than data volume

Dengan rasio memori terhadap data sekitar 7:1, sebagian besar permintaan dilayani dari memori.

Instance type	YCSB configuration	Workload	QPS	Average latency (µs)	99th percentile latency (µs)
`tair.localssd.c1m8.xlarge`	recordcount=20.000.000 run_operationcount=20.000.000 threads=64 fieldlength=100	Load	42.275	1.511	4.987
		Uniform-Read	104.303	611	2.163
		Zipfian-Read	105.217	605	2.259
		Uniform-50%Read-50%Update	52.267	Read: 1.114	Read: 4.191
		Uniform-50%Read-50%Update	52.267	Update: 1.329	Update: 4.827
`tair.localssd.c1m8.2xlarge`	recordcount=40.000.000 run_operationcount=40.000.000 threads=128 fieldlength=100	Load	64.104	1.993	5.951
		Uniform-Read	174.533	730	2.049
		Zipfian-Read	177.612	718	2.033
		Uniform-50%Read-50%Update	75.762	Read: 1.509	Read: 4.667
		Uniform-50%Read-50%Update	75.762	Update: 1.864	Update: 5.479
`tair.localssd.c1m8.with.proxy.xlarge (3 shards)`	recordcount=60.000.000 run_operationcount=60.000.000 threads=192 fieldlength=100	Load	116.107	1.647	9.983
		Uniform-Read	225.613	847	3.687
		Zipfian-Read	234.014	816	3.279
		Uniform-50%Read-50%Update	138.884	Read: 1.321	Read: 8.131
		Uniform-50%Read-50%Update	138.884	Update: 1.435	Update: 8.479
`tair.localssd.c1m8.with.proxy.2xlarge (3 shards)`	recordcount=120.000.000 run_operationcount=120.000.000 threads=384 fieldlength=100	Load	170.447	2.246	11.935
		Uniform-Read	323.136	1.184	4.005
		Zipfian-Read	330.023	1.159	3.951
		Uniform-50%Read-50%Update	207.350	Read: 1.759	Read: 9.031
		Uniform-50%Read-50%Update	207.350	Update: 1.934	Update: 9.439

Data volume larger than memory size

Dengan rasio memori terhadap data sekitar 1:4, sebagian besar permintaan memerlukan I/O disk, sehingga latensi lebih tinggi dibandingkan skenario dominasi memori di atas.

Instance type	YCSB configuration	Workload	QPS	Average latency (µs)	99th percentile latency (µs)
`tair.localssd.c1m8.xlarge`	recordcount=640.000.000 run_operationcount=640.000.000 threads=64 fieldlength=100	Load	33.198	1.925	7.699
		Uniform-Read	48.243	1.324	3.659
		Zipfian-Read	66.635	958	3.003
		Uniform-50%Read-50%Update	28.779	Read: 2.169	Read: 7.531
		Uniform-50%Read-50%Update	28.779	Update: 2.272	Update: 7.851
`tair.localssd.c1m8.2xlarge`	recordcount=1.280.000.000 run_operationcount=1.280.000.000 threads=128 fieldlength=100	Load	48.605	2.631	8.463
		Uniform-Read	83.687	1.527	3.889
		Zipfian-Read	113.742	1.123	3.351
		Uniform-50%Read-50%Update	46.606	Read: 2.652	Read: 8.815
		Uniform-50%Read-50%Update	46.606	Update: 2.835	Update: 9.367
`tair.localssd.c1m8.with.proxy.xlarge (3 shards)`	recordcount=1.920.000.000 run_operationcount=1.920.000.000 threads=192 fieldlength=100	Load	89.421	2.144	15.351
		Uniform-Read	115.967	1.653	9.311
		Zipfian-Read	155.686	1.231	6.359
		Uniform-50%Read-50%Update	72.890	Read: 2.585	Read: 16.479
		Uniform-50%Read-50%Update	72.890	Update: 2.676	Update: 16.719
`tair.localssd.c1m8.with.proxy.2xlarge (3 shards)`	recordcount=3.840.000.000 run_operationcount=3.840.000.000 threads=384 fieldlength=100	Load	137.521	2.789	16.591
		Uniform-Read	185.273	2.070	8.575
		Zipfian-Read	238.099	1.610	6.203
		Uniform-50%Read-50%Update	128.080	Read: 2.930	Read: 15.535
		Uniform-50%Read-50%Update	128.080	Update: 3.060	Update: 15.807

Performance comparison chart

References

ESSD/SSD-based instances