ext是Linux原生的文件系统，版本依次从ext2迭代到ext4，xfs文件系统在90年代以商业化的形式出现，2009年之后才移植到Linux平台，从特征上来说，ext以轻量稳定为核心，xfs以大容量高性能为核心，xfs是否一定优于ext4，本文就该问题进行一个简单的测试验证。
本文的重点在于对比笔者自己常用的场景下，ext4和xfs在“相同条件下”的性能表现，而非测试硬件或者软件自身的极限性能。
fio是Linux下的一个IO压力测试工具，可以生成自定义 IO 负载，模拟业务读写模型，测量存储性能，创建多个线程 / 进程，按照配置持续发起 IO 请求；统计 带宽 (BW)、IOPS、延迟分布、队列利用率 util等一系列跟性能相关的指标。
本文测试环境为ubuntu 24操作系统的虚拟机，物理磁盘是STAT类型的SSD，虚拟机的配置为4C+8G的配置，如下，分别挂载ext4和xfs两块文件类型的磁盘，用以做IO压力对比测试。
root@ubuntu30:/home/ubuntu# lsblk -f
NAME FSTYPE FSVER LABEL UUID FSAVAIL FSUSE% MOUNTPOINTS
sda
├─sda1
├─sda2 ext4 1.0 f24e6ace-4e10-407d-9da0-300792128569 802.2M 11% /boot
└─sda3 xfs 14609fa0-73be-4194-a72c-1672aa73e953 41.3G 30% /
sdb ext4 1.0 be8519af-ce4a-4860-83a7-429b864fe1d1 45.7G 17% /usr/data
sr0 iso9660 Joliet Extension Ubuntu-Server 24.04.3 LTS amd64 2025-08-05-23-54-07-00
root@ubuntu30:/home/ubuntu#
root@ubuntu30:/home/ubuntu#对于磁盘的测试，这里使用fio命令，用随机读加写，70%的读操作，30%的写操作，队列深度为32的测试方式
### ext4磁盘分区
fio -name=randrw-8k -ioengine=libaio -direct=1 -rw=randrw -rwmixread=70 -bs=8k -size=10G -numjobs=4 -runtime=120 --time_based -iodepth=32 -filename=/fio.test
### xfs磁盘分区
fio -name=randrw-8k -ioengine=libaio -direct=1 -rw=randrw -rwmixread=70 -bs=8k -size=10G -numjobs=4 -runtime=120 --time_based -iodepth=32 -filename=/usr/data/fio.test一次典型的fio测试如下
root@ubuntu30:/home/ubuntu# fio -name=randrw-8k -ioengine=libaio -direct=1 -rw=randrw -rwmixread=70 -bs=8k -size=10G -numjobs=4 -runtime=120 --time_based -iodepth=32 -filename=/usr/data/fio.test
randrw-8k: (g=0): rw=randrw, bs=(R) 8192B-8192B, (W) 8192B-8192B, (T) 8192B-8192B, ioengine=libaio, iodepth=32
...
fio-3.36
Starting 4 processes
Jobs: 4 (f=4): [m(4)][100.0%][r=87.9MiB/s,w=38.3MiB/s][r=11.3k,w=4899 IOPS][eta 00m:00s]
randrw-8k: (groupid=0, jobs=1): err= 0: pid=1547: Wed Jul 15 08:01:29 2026
read: IOPS=3390, BW=26.5MiB/s (27.8MB/s)(3179MiB/120009msec)
slat (nsec): min=1428, max=169935k, avg=69690.46, stdev=670894.67
clat (usec): min=2, max=204221, avg=6218.01, stdev=6328.43
lat (usec): min=58, max=204756, avg=6287.70, stdev=6357.47
clat percentiles (usec):
| 1.00th=[ 453], 5.00th=[ 857], 10.00th=[ 1205], 20.00th=[ 1827],
| 30.00th=[ 2507], 40.00th=[ 3359], 50.00th=[ 4424], 60.00th=[ 5735],
| 70.00th=[ 7242], 80.00th=[ 9372], 90.00th=[ 13173], 95.00th=[ 17171],
| 99.00th=[ 29230], 99.50th=[ 36963], 99.90th=[ 57934], 99.95th=[ 71828],
| 99.99th=[109577]
bw ( KiB/s): min=11328, max=48176, per=25.01%, avg=27111.31, stdev=6098.33, samples=238
iops : min= 1416, max= 6022, avg=3388.67, stdev=762.29, samples=238
write: IOPS=1456, BW=11.4MiB/s (11.9MB/s)(1365MiB/120009msec); 0 zone resets
slat (nsec): min=1559, max=57787k, avg=72491.01, stdev=582537.80
clat (nsec): min=1158, max=204767k, avg=7252447.95, stdev=7918867.24
lat (usec): min=78, max=204771, avg=7324.94, stdev=7932.98
clat percentiles (usec):
| 1.00th=[ 441], 5.00th=[ 832], 10.00th=[ 1221], 20.00th=[ 1942],
| 30.00th=[ 2737], 40.00th=[ 3687], 50.00th=[ 4883], 60.00th=[ 6456],
| 70.00th=[ 8356], 80.00th=[ 10945], 90.00th=[ 15664], 95.00th=[ 20841],
| 99.00th=[ 36963], 99.50th=[ 45876], 99.90th=[ 72877], 99.95th=[ 86508],
| 99.99th=[126354]
bw ( KiB/s): min= 4624, max=21008, per=24.98%, avg=11646.62, stdev=2694.12, samples=238
iops : min= 578, max= 2626, avg=1455.59, stdev=336.76, samples=238
lat (usec) : 2=0.01%, 4=0.01%, 10=0.01%, 20=0.01%, 50=0.02%
lat (usec) : 100=0.05%, 250=0.20%, 500=1.07%, 750=2.42%, 1000=3.30%
lat (msec) : 2=15.01%, 4=23.20%, 10=35.21%, 20=15.62%, 50=3.67%
lat (msec) : 100=0.21%, 250=0.02%
cpu : usr=1.28%, sys=28.10%, ctx=208596, majf=0, minf=17
IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=100.0%, >=64=0.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.1%, 64=0.0%, >=64=0.0%
issued rwts: total=406867,174750,0,0 short=0,0,0,0 dropped=0,0,0,0
latency : target=0, window=0, percentile=100.00%, depth=32
randrw-8k: (groupid=0, jobs=1): err= 0: pid=1548: Wed Jul 15 08:01:29 2026
read: IOPS=3387, BW=26.5MiB/s (27.7MB/s)(3176MiB/120009msec)
slat (nsec): min=1454, max=152785k, avg=69838.54, stdev=700426.96
clat (nsec): min=1947, max=175067k, avg=6208002.91, stdev=6417046.59
lat (usec): min=70, max=175070, avg=6277.84, stdev=6444.81
clat percentiles (usec):
| 1.00th=[ 461], 5.00th=[ 873], 10.00th=[ 1221], 20.00th=[ 1844],
| 30.00th=[ 2540], 40.00th=[ 3359], 50.00th=[ 4424], 60.00th=[ 5669],
| 70.00th=[ 7242], 80.00th=[ 9372], 90.00th=[ 13042], 95.00th=[ 16909],
| 99.00th=[ 29230], 99.50th=[ 36963], 99.90th=[ 62129], 99.95th=[ 79168],
| 99.99th=[132645]
bw ( KiB/s): min=10506, max=47075, per=24.99%, avg=27087.85, stdev=6101.88, samples=238
iops : min= 1313, max= 5884, avg=3385.75, stdev=762.74, samples=238
write: IOPS=1460, BW=11.4MiB/s (12.0MB/s)(1369MiB/120009msec); 0 zone resets
slat (nsec): min=1569, max=69377k, avg=73472.05, stdev=628177.68
clat (usec): min=11, max=165134, avg=7268.70, stdev=7838.10
lat (usec): min=90, max=165139, avg=7342.17, stdev=7852.69
clat percentiles (usec):
| 1.00th=[ 449], 5.00th=[ 848], 10.00th=[ 1237], 20.00th=[ 1975],
| 30.00th=[ 2769], 40.00th=[ 3752], 50.00th=[ 5014], 60.00th=[ 6521],
| 70.00th=[ 8356], 80.00th=[ 11076], 90.00th=[ 15664], 95.00th=[ 20841],
| 99.00th=[ 36439], 99.50th=[ 45876], 99.90th=[ 72877], 99.95th=[ 88605],
| 99.99th=[145753]
bw ( KiB/s): min= 4614, max=20254, per=25.05%, avg=11677.46, stdev=2682.05, samples=238
iops : min= 576, max= 2531, avg=1459.43, stdev=335.24, samples=238
lat (usec) : 2=0.01%, 4=0.01%, 10=0.01%, 20=0.01%, 50=0.02%
lat (usec) : 100=0.05%, 250=0.18%, 500=1.04%, 750=2.33%, 1000=3.21%
lat (msec) : 2=14.97%, 4=23.24%, 10=35.57%, 20=15.54%, 50=3.59%
lat (msec) : 100=0.23%, 250=0.02%
cpu : usr=1.22%, sys=28.38%, ctx=207948, majf=0, minf=17
IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=100.0%, >=64=0.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.1%, 64=0.0%, >=64=0.0%
issued rwts: total=406499,175255,0,0 short=0,0,0,0 dropped=0,0,0,0
latency : target=0, window=0, percentile=100.00%, depth=32
randrw-8k: (groupid=0, jobs=1): err= 0: pid=1549: Wed Jul 15 08:01:29 2026
read: IOPS=3417, BW=26.7MiB/s (28.0MB/s)(3204MiB/120008msec)
slat (nsec): min=1460, max=120727k, avg=67489.07, stdev=586036.62
clat (usec): min=2, max=161462, avg=6152.00, stdev=6311.85
lat (usec): min=83, max=161465, avg=6219.49, stdev=6333.94
clat percentiles (usec):
| 1.00th=[ 461], 5.00th=[ 873], 10.00th=[ 1205], 20.00th=[ 1844],
| 30.00th=[ 2507], 40.00th=[ 3326], 50.00th=[ 4359], 60.00th=[ 5669],
| 70.00th=[ 7177], 80.00th=[ 9372], 90.00th=[ 12911], 95.00th=[ 16909],
| 99.00th=[ 28705], 99.50th=[ 35914], 99.90th=[ 59507], 99.95th=[ 74974],
| 99.99th=[135267]
bw ( KiB/s): min= 9992, max=51424, per=25.20%, avg=27315.33, stdev=6352.85, samples=238
iops : min= 1249, max= 6428, avg=3414.19, stdev=794.11, samples=238
write: IOPS=1465, BW=11.5MiB/s (12.0MB/s)(1374MiB/120008msec); 0 zone resets
slat (nsec): min=1596, max=72174k, avg=73942.54, stdev=643502.98
clat (nsec): min=1791, max=152333k, avg=7247229.63, stdev=7872188.82
lat (usec): min=105, max=152336, avg=7321.17, stdev=7896.87
clat percentiles (usec):
| 1.00th=[ 445], 5.00th=[ 857], 10.00th=[ 1254], 20.00th=[ 1975],
| 30.00th=[ 2769], 40.00th=[ 3720], 50.00th=[ 4948], 60.00th=[ 6456],
| 70.00th=[ 8356], 80.00th=[ 10945], 90.00th=[ 15533], 95.00th=[ 20841],
| 99.00th=[ 35914], 99.50th=[ 44827], 99.90th=[ 76022], 99.95th=[102237],
| 99.99th=[141558]
bw ( KiB/s): min= 4318, max=20733, per=25.14%, avg=11719.82, stdev=2678.11, samples=238
iops : min= 539, max= 2591, avg=1464.76, stdev=334.77, samples=238
lat (usec) : 2=0.01%, 4=0.01%, 10=0.01%, 20=0.01%, 50=0.02%
lat (usec) : 100=0.04%, 250=0.19%, 500=1.03%, 750=2.33%, 1000=3.21%
lat (msec) : 2=15.01%, 4=23.56%, 10=35.39%, 20=15.45%, 50=3.55%
lat (msec) : 100=0.18%, 250=0.03%
cpu : usr=1.28%, sys=27.71%, ctx=208372, majf=0, minf=17
IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=100.0%, >=64=0.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.1%, 64=0.0%, >=64=0.0%
issued rwts: total=410091,175924,0,0 short=0,0,0,0 dropped=0,0,0,0
latency : target=0, window=0, percentile=100.00%, depth=32
randrw-8k: (groupid=0, jobs=1): err= 0: pid=1550: Wed Jul 15 08:01:29 2026
read: IOPS=3355, BW=26.2MiB/s (27.5MB/s)(3146MiB/120008msec)
slat (nsec): min=1458, max=64036k, avg=69779.03, stdev=556041.90
clat (nsec): min=1061, max=146162k, avg=6257973.28, stdev=6147059.49
lat (usec): min=84, max=146170, avg=6327.75, stdev=6160.76
clat percentiles (usec):
| 1.00th=[ 465], 5.00th=[ 889], 10.00th=[ 1237], 20.00th=[ 1893],
| 30.00th=[ 2606], 40.00th=[ 3458], 50.00th=[ 4490], 60.00th=[ 5735],
| 70.00th=[ 7308], 80.00th=[ 9503], 90.00th=[13173], 95.00th=[17171],
| 99.00th=[29492], 99.50th=[36439], 99.90th=[53740], 99.95th=[63177],
| 99.99th=[92799]
bw ( KiB/s): min=11344, max=42874, per=24.80%, avg=26885.74, stdev=5860.63, samples=238
iops : min= 1418, max= 5359, avg=3360.49, stdev=732.59, samples=238
write: IOPS=1444, BW=11.3MiB/s (11.8MB/s)(1355MiB/120008msec); 0 zone resets
slat (nsec): min=1558, max=59757k, avg=74464.14, stdev=577719.22
clat (usec): min=2, max=190257, avg=7366.86, stdev=7702.63
lat (usec): min=78, max=190259, avg=7441.33, stdev=7721.18
clat percentiles (usec):
| 1.00th=[ 449], 5.00th=[ 873], 10.00th=[ 1287], 20.00th=[ 2040],
| 30.00th=[ 2835], 40.00th=[ 3851], 50.00th=[ 5080], 60.00th=[ 6587],
| 70.00th=[ 8455], 80.00th=[ 11207], 90.00th=[ 15795], 95.00th=[ 21103],
| 99.00th=[ 36963], 99.50th=[ 45351], 99.90th=[ 69731], 99.95th=[ 76022],
| 99.99th=[115868]
bw ( KiB/s): min= 5360, max=19120, per=24.84%, avg=11578.18, stdev=2567.29, samples=238
iops : min= 670, max= 2390, avg=1447.01, stdev=320.91, samples=238
lat (usec) : 2=0.01%, 4=0.01%, 10=0.01%, 20=0.01%, 50=0.02%
lat (usec) : 100=0.05%, 250=0.19%, 500=0.98%, 750=2.23%, 1000=3.04%
lat (msec) : 2=14.43%, 4=23.33%, 10=35.97%, 20=15.80%, 50=3.75%
lat (msec) : 100=0.20%, 250=0.01%
cpu : usr=1.22%, sys=28.24%, ctx=201096, majf=0, minf=17
IO depths : 1=0.1%, 2=0.1%, 4=0.1%, 8=0.1%, 16=0.1%, 32=100.0%, >=64=0.0%
submit : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.0%, 64=0.0%, >=64=0.0%
complete : 0=0.0%, 4=100.0%, 8=0.0%, 16=0.0%, 32=0.1%, 64=0.0%, >=64=0.0%
issued rwts: total=402719,173382,0,0 short=0,0,0,0 dropped=0,0,0,0
latency : target=0, window=0, percentile=100.00%, depth=32
Run status group 0 (all jobs):
READ: bw=106MiB/s (111MB/s), 26.2MiB/s-26.7MiB/s (27.5MB/s-28.0MB/s), io=12.4GiB (13.3GB), run=120008-120009msec
WRITE: bw=45.5MiB/s (47.7MB/s), 11.3MiB/s-11.5MiB/s (11.8MB/s-12.0MB/s), io=5463MiB (5729MB), run=120008-120009msec
Disk stats (read/write):
sdb: ios=1625786/699203, sectors=26014920/11187536, merge=146/54, ticks=8435438/4194989, in_queue=12630428, util=100.00%
root@ubuntu30:/home/ubuntu#
root@ubuntu30:/home/ubuntu#
root@ubuntu30:/home/ubuntu#这里采用shell循环交替测试的方式，对于ext4和xfs分别测试10次，最终把测试结果丢给AI，让AI帮助汇总测试结果并生成相关图例
#!/bin/bash
# ============================================
# fio 自动化测试
# EXT4 / XFS 各 10 次
# ============================================
EXT4_FILE="/fio.test"
XFS_FILE="/usr/data/fio.test"
RESULT_DIR="./"
EXT4_LOG="${RESULT_DIR}/ext4_results.log"
XFS_LOG="${RESULT_DIR}/xfs_results.log"
SUMMARY_LOG="${RESULT_DIR}/summary.log"
RUNTIME="120"
NUM_JOBS="4"
IODEPTH="32"
BS="8k"
SIZE="10G"
RWMIXREAD="70"
mkdir -p "${RESULT_DIR}"
# 初始化 summary
if [ ! -f "${SUMMARY_LOG}" ]; then
echo "========================================" > "${SUMMARY_LOG}"
echo "fio benchmark (EN locale)" >> "${SUMMARY_LOG}"
echo "Start: $(date '+%Y-%m-%d %H:%M:%S')" >> "${SUMMARY_LOG}"
echo "========================================" >> "${SUMMARY_LOG}"
fi
for i in $(seq 1 10); do
echo ""
echo ">>>> Round ${i} start: $(date '+%H:%M:%S')"
echo "===== Round ${i} | $(date '+%Y-%m-%d %H:%M:%S') =====" >> "${EXT4_LOG}"
echo "===== Round ${i} | $(date '+%Y-%m-%d %H:%M:%S') =====" >> "${XFS_LOG}"
# ---------- EXT4 ----------
echo " -> Running EXT4..."
LC_ALL=C fio \
-name=randrw-8k \
-ioengine=libaio \
-direct=1 \
-rw=randrw \
-rwmixread=${RWMIXREAD} \
-bs=${BS} \
-size=${SIZE} \
-numjobs=${NUM_JOBS} \
-runtime=${RUNTIME} \
--time_based \
-iodepth=${IODEPTH} \
-filename=${EXT4_FILE} \
>> "${EXT4_LOG}" 2>&1
echo " -> EXT4 done"
sleep 5
# ---------- XFS ----------
echo " -> Running XFS..."
LC_ALL=C fio \
-name=randrw-8k \
-ioengine=libaio \
-direct=1 \
-rw=randrw \
-rwmixread=${RWMIXREAD} \
-bs=${BS} \
-size=${SIZE} \
-numjobs=${NUM_JOBS} \
-runtime=${RUNTIME} \
--time_based \
-iodepth=${IODEPTH} \
-filename=${XFS_FILE} \
>> "${XFS_LOG}" 2>&1
echo " -> XFS done"
echo "Round ${i}: EXT4=$(date '+%H:%M:%S') | XFS=$(date '+%H:%M:%S')" >> "${SUMMARY_LOG}"
echo "" >> "${EXT4_LOG}"
echo "" >> "${XFS_LOG}"
if [ ${i} -lt 10 ]; then
echo " -> Sleep 5s..."
sleep 5
fi
done
echo "========================================" >> "${SUMMARY_LOG}"
echo "Finished: $(date '+%Y-%m-%d %H:%M:%S')" >> "${SUMMARY_LOG}"
echo "========================================" >> "${SUMMARY_LOG}"
echo ""
echo "All tests finished"
echo "EXT4 log: ${EXT4_LOG}"
echo "XFS log: ${XFS_LOG}"这里重点绘制出来两个核心技术指标IOPS和读写IO延迟比，从中可以发现
1，IOPS这一指标，分别为read IOPS和write IOPS，整体上看两者没有明显的差异，相差不到1%，但是ext4相对xfs来说波动更小
2，对于read clat 95以及clat 99的percentiles指标，两者几乎一致，相差同样小于1%
以下图例数据来自于shell脚本的生成的fio测试结果的汇总，图例由AI根据测试结果整理生成
这里安装了2个实例，分别是postgresql9800和postgresql9900，两个实例的数据文件分别位于ext4和xfs所在的磁盘分区，两个postgresql实例配置完全一致，其中share_buffer配置为1G。
root@ubuntu30:/usr/local/pg_install# systemctl status postgresql9800
● postgresql9800.service - PostgreSQL database server
Loaded: loaded (/etc/systemd/system/postgresql9800.service; enabled; preset: enabled)
Active: active (running) since Thu 2026-07-16 04:54:33 UTC; 1h 12min ago
Process: 36462 ExecStart=/usr/local/pgsql18/server/bin/pg_ctl start -D ${PGDATA} -s -w -t ${PGSTARTTIMEOUT} (code=exited, status=0/SUCCESS)
Main PID: 36465 (postgres)
Tasks: 10 (limit: 9374)
Memory: 6.1G (peak: 6.8G swap: 680.0K swap peak: 680.0K)
CPU: 10min 47.010s
CGroup: /system.slice/postgresql9800.service
├─36465 /usr/local/pgsql18/server/bin/postgres -D /usr/local/pgsql18/pg9800/data
├─36466 "postgres: logger "
├─36467 "postgres: io worker 0"
├─36468 "postgres: io worker 1"
├─36469 "postgres: io worker 2"
├─36470 "postgres: checkpointer "
├─36471 "postgres: background writer "
├─36473 "postgres: walwriter "
├─36474 "postgres: autovacuum launcher "
└─36475 "postgres: logical replication launcher "
Jul 16 04:54:33 ubuntu30 systemd[1]: Starting postgresql9800.service - PostgreSQL database server...
Jul 16 04:54:33 ubuntu30 pg_ctl[36465]: 2026-07-16 04:54:33.406 UTC [36465] LOG: redirecting log output to logging collector process
Jul 16 04:54:33 ubuntu30 pg_ctl[36465]: 2026-07-16 04:54:33.406 UTC [36465] HINT: Future log output will appear in directory "/usr/local/pgsql18/pg9800/log".
Jul 16 04:54:33 ubuntu30 systemd[1]: Started postgresql9800.service - PostgreSQL database server.
root@ubuntu30:/usr/local/pg_install# systemctl status postgresql9900
● postgresql9900.service - PostgreSQL database server
Loaded: loaded (/etc/systemd/system/postgresql9900.service; enabled; preset: enabled)
Active: active (running) since Thu 2026-07-16 06:07:12 UTC; 16s ago
Process: 37352 ExecStart=/usr/data/pgsql18/server/bin/pg_ctl start -D ${PGDATA} -s -w -t ${PGSTARTTIMEOUT} (code=exited, status=0/SUCCESS)
Main PID: 37355 (postgres)
Tasks: 10 (limit: 9374)
Memory: 57.1M (peak: 57.6M)
CPU: 82ms
CGroup: /system.slice/postgresql9900.service
├─37355 /usr/data/pgsql18/server/bin/postgres -D /usr/data/pgsql18/pg9900/data
├─37356 "postgres: logger "
├─37357 "postgres: io worker 0"
├─37358 "postgres: io worker 1"
├─37359 "postgres: io worker 2"
├─37360 "postgres: checkpointer "
├─37361 "postgres: background writer "
├─37363 "postgres: walwriter "
├─37364 "postgres: autovacuum launcher "
└─37365 "postgres: logical replication launcher "
Jul 16 06:07:12 ubuntu30 systemd[1]: Starting postgresql9900.service - PostgreSQL database server...
Jul 16 06:07:12 ubuntu30 pg_ctl[37355]: 2026-07-16 06:07:12.896 UTC [37355] LOG: redirecting log output to logging collector process
Jul 16 06:07:12 ubuntu30 pg_ctl[37355]: 2026-07-16 06:07:12.896 UTC [37355] HINT: Future log output will appear in directory "/usr/data/pgsql18/pg9900/log".
Jul 16 06:07:12 ubuntu30 systemd[1]: Started postgresql9900.service - PostgreSQL database server.
root@ubuntu30:/usr/local/pg_install#这里基于pgbench创建一个50000000条数据的测试环境（pgbench -h 127.0.0.1 -p 9800 -U postgres -d pgbench_db -i -s 500），两个实例的生成的测试数据完全一致。
root@ubuntu30:/usr/local/pg_install# psql -h 127.0.0.1 -p 9800 -U postgres postgres
Password for user postgres:
psql (18.3)
Type "help" for help.
postgres=# create database pgbench_db;
CREATE DATABASE
postgres=# \q
root@ubuntu30:/usr/local/pg_install#
root@ubuntu30:/usr/local/pg_install# pgbench -h 127.0.0.1 -U postgres -p 9800 pgbench_db -i
Password:
dropping old tables...
NOTICE: table "pgbench_accounts" does not exist, skipping
NOTICE: table "pgbench_branches" does not exist, skipping
NOTICE: table "pgbench_history" does not exist, skipping
NOTICE: table "pgbench_tellers" does not exist, skipping
creating tables...
generating data (client-side)...
vacuuming...
creating primary keys...
done in 0.25 s (drop tables 0.00 s, create tables 0.00 s, client-side generate 0.10 s, vacuum 0.04 s, primary keys 0.10 s).
root@ubuntu30:/usr/local/pg_install#
root@ubuntu30:/usr/local/pg_install# pgbench -h 127.0.0.1 -p 9800 -U postgres -d pgbench_db -i -s 500
Password:
dropping old tables...
creating tables...
generating data (client-side)...
vacuuming...
creating primary keys...
done in 169.31 s (drop tables 0.03 s, create tables 0.01 s, client-side generate 100.66 s, vacuum 3.76 s, primary keys 64.86 s).
root@ubuntu30:/usr/local/pg_install#利用如下脚本，分别用"2 4 8 16 32 64" 线程数对两个postgresql分别进行压力测试，每个一组测试运行60秒，测试期间会关停另一个实例的服务，防止两个实例之间相互干扰。
HOST=127.0.0.1
PGPORT=9800
DBNAME=pgbench_db
DBUSER=postgres
export PGPASSWORD=******
RUNSEC=60 # 运行时间
THREADS="2 4 8 16 32 64" 线程数
for threadCnt in $THREADS ; do
tps=`pgbench -r -T$RUNSEC $DBNAME -U$DBUSER -h $HOST -p $PGPORT -c$threadCnt -j$threadCnt 2>&1 |awk '/tps/{print $3}'`
echo "concurrent: $threadCnt ,postgres tps: ${tps:-0}"
done测试结果
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9800_test.sh
concurrent: 2 ,postgres tps: 873.224907
concurrent: 4 ,postgres tps: 1150.755581
concurrent: 8 ,postgres tps: 2161.466191
concurrent: 16 ,postgres tps: 3119.279227
concurrent: 32 ,postgres tps: 3055.131493
concurrent: 64 ,postgres tps: 2835.200960
concurrent: 128 ,postgres tps: 2859.829723
root@ubuntu30:/usr/local/pg_install#
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9800_test.sh
concurrent: 2 ,postgres tps: 743.739950
concurrent: 4 ,postgres tps: 1066.351943
concurrent: 8 ,postgres tps: 2234.594472
concurrent: 16 ,postgres tps: 3165.805190
concurrent: 32 ,postgres tps: 3402.018566
concurrent: 64 ,postgres tps: 3046.540664
concurrent: 128 ,postgres tps: 2965.726659
root@ubuntu30:/usr/local/pg_install#
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9800_test.sh
concurrent: 2 ,postgres tps: 761.140493
concurrent: 4 ,postgres tps: 1353.549987
concurrent: 8 ,postgres tps: 2392.809218
concurrent: 16 ,postgres tps: 3337.523674
concurrent: 32 ,postgres tps: 3223.437427
concurrent: 64 ,postgres tps: 3057.000292
concurrent: 128 ,postgres tps: 2956.257316
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9800_test.sh
concurrent: 2 ,postgres tps: 809.081475
concurrent: 4 ,postgres tps: 1218.683222
concurrent: 8 ,postgres tps: 2411.138366
concurrent: 16 ,postgres tps: 3348.417863
concurrent: 32 ,postgres tps: 3420.398202
concurrent: 64 ,postgres tps: 3270.371034
concurrent: 128 ,postgres tps: 3073.792106
root@ubuntu30:/usr/local/pg_install#
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9800_test.sh
concurrent: 2 ,postgres tps: 847.993710
concurrent: 4 ,postgres tps: 1205.544455
concurrent: 8 ,postgres tps: 2299.767415
concurrent: 16 ,postgres tps: 3416.636078
concurrent: 32 ,postgres tps: 3197.307265
concurrent: 64 ,postgres tps: 3231.308020
concurrent: 128 ,postgres tps: 3262.546457
root@ubuntu30:/usr/local/pg_install#
root@ubuntu30:/usr/local/pg_install#
root@ubuntu30:/usr/local/pg_install# systemctl stop postgresql9800
root@ubuntu30:/usr/local/pg_install#
root@ubuntu30:/usr/local/pg_install# systemctl start postgresql9900
root@ubuntu30:/usr/local/pg_install#
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9900_test.sh
concurrent: 2 ,postgres tps: 910.148271
concurrent: 4 ,postgres tps: 945.297788
concurrent: 8 ,postgres tps: 1984.848836
concurrent: 16 ,postgres tps: 3463.093250
concurrent: 32 ,postgres tps: 3415.695770
concurrent: 64 ,postgres tps: 3299.900340
concurrent: 128 ,postgres tps: 3163.273326
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9900_test.sh
concurrent: 2 ,postgres tps: 972.202275
concurrent: 4 ,postgres tps: 1199.432083
concurrent: 8 ,postgres tps: 2331.193031
concurrent: 16 ,postgres tps: 3409.552122
concurrent: 32 ,postgres tps: 3447.767055
concurrent: 64 ,postgres tps: 3356.412922
concurrent: 128 ,postgres tps: 3315.727438
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9900_test.sh
concurrent: 2 ,postgres tps: 878.067693
concurrent: 4 ,postgres tps: 1291.111674
concurrent: 8 ,postgres tps: 2399.937215
concurrent: 16 ,postgres tps: 3561.428644
concurrent: 32 ,postgres tps: 3711.227386
concurrent: 64 ,postgres tps: 3414.528734
concurrent: 128 ,postgres tps: 3197.026801
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9900_test.sh
concurrent: 2 ,postgres tps: 918.756230
concurrent: 4 ,postgres tps: 1202.624052
concurrent: 8 ,postgres tps: 2359.952643
concurrent: 16 ,postgres tps: 3539.555993
concurrent: 32 ,postgres tps: 3560.534475
concurrent: 64 ,postgres tps: 3462.045914
concurrent: 128 ,postgres tps: 2837.921359
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9900_test.sh
concurrent: 2 ,postgres tps: 791.815163
concurrent: 4 ,postgres tps: 1205.374922
concurrent: 8 ,postgres tps: 2493.116099
concurrent: 16 ,postgres tps: 3456.360421
concurrent: 32 ,postgres tps: 3568.589185
concurrent: 64 ,postgres tps: 3331.357654
concurrent: 128 ,postgres tps: 3262.971818
root@ubuntu30:/usr/local/pg_install# bash pgbench_postgresql9900_test.sh
concurrent: 2 ,postgres tps: 825.601480
concurrent: 4 ,postgres tps: 1210.087214
concurrent: 8 ,postgres tps: 2508.689554
concurrent: 16 ,postgres tps: 3591.199202
concurrent: 32 ,postgres tps: 3548.136655
concurrent: 64 ,postgres tps: 3440.649905
concurrent: 128 ,postgres tps: 3115.197624
root@ubuntu30:/usr/local/pg_install#如下图是上述两次测试结果，交给AI后生成的折线图，可以看出
在中低并发（2~8线程）：ext4 与 XFS的TPS差别很小
在中高并发（16~128线程）：两者性能非常接近，XFS比ext4高出约10%左右，并且tps更加趋于稳定
平均 TPS 数据汇总
|
并发数 |
ext4 平均 TPS |
ext4 标准差 |
xfs 平均 TPS |
xfs 标准差 |
|---|---|---|---|---|
|
2 |
807.0 |
49.3 |
882.8 |
60.0 |
|
4 |
1199.0 |
94.1 |
1175.7 |
107.8 |
|
8 |
2300.0 |
94.3 |
2346.3 |
174.1 |
|
16 |
3277.5 |
114.5 |
3503.5 |
64.6 |
|
32 |
3259.7 |
136.5 |
3542.0 |
95.4 |
|
64 |
3088.1 |
155.2 |
3384.2 |
58.9 |
|
128 |
3023.6 |
137.3 |
3148.7 |
153.4 |
fio 磁盘 IO 测试中：
1，IOPS方面，ext4 与 XFS 差异 < 1%，ext4 波动更小
2，读写延迟（clat 95 / clat 99）方面，两者几乎一致，差异 < 1%
在ext4与xfs上的postgresql实例测试中：
1，在中低并发下ext4略有优势，
2，而中高并发下xfs表现出来一定的优势，本文测试中大概有10%左右的提升
鉴于ext4与xfs都是非常成熟的文件系统，都具经历了长期的考验，仅通过本文相对简单的测试，说明在本文的测试场景下，并没有发现两者有明显的差异，而不是两者没有差异。
相关链接，本文没有参考该链接的测试，只是觉得测试的很全面，放在这里
https://www.enterprisedb.com/blog/postgres-vs-file-systems-performance-comparison
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。