fio.md

This page represents a collection of fio performance tests, tuned for a Kubernetes etcd workload per this blog post, against various storage and platforms.

The goal is to execute the below fio command on as many different places as possible to gauge relative performance.

fio --rw=write --ioengine=sync --fdatasync=1 --directory=test-data --size=22m --bs=2300 --name=mytest

These tests are completely unscientific and only serve to provide a sampling for anecdotal comparisons.

RHEL 7 - NVMe + VDO

This is a physical node running RHEL 7.8. The underlying storage is a consumer NVMe device with a VDO pool created on top. VDO is doing deduplication but not compression.

mytest: (g=0): rw=write, bs=(R) 2300B-2300B, (W) 2300B-2300B, (T) 2300B-2300B, ioengine=sync, iodepth=1
fio-3.7
Starting 1 process
mytest: Laying out IO file (1 file / 22MiB)
Jobs: 1 (f=1): [W(1)][100.0%][r=0KiB/s,w=1498KiB/s][r=0,w=667 IOPS][eta 00m:00s]
mytest: (groupid=0, jobs=1): err= 0: pid=9922: Thu Oct 29 14:44:39 2020
  write: IOPS=664, BW=1493KiB/s (1529kB/s)(21.0MiB/15089msec)
    clat (nsec): min=4295, max=43404, avg=8670.40, stdev=2712.21
     lat (nsec): min=4500, max=44118, avg=8874.89, stdev=2735.94
    clat percentiles (nsec):
     |  1.00th=[ 5280],  5.00th=[ 5664], 10.00th=[ 5856], 20.00th=[ 6240],
     | 30.00th=[ 6816], 40.00th=[ 7776], 50.00th=[ 8512], 60.00th=[ 8896],
     | 70.00th=[ 9536], 80.00th=[10176], 90.00th=[11968], 95.00th=[13504],
     | 99.00th=[18304], 99.50th=[20352], 99.90th=[24704], 99.95th=[25728],
     | 99.99th=[32128]
   bw (  KiB/s): min= 1455, max= 1522, per=100.00%, avg=1492.43, stdev=19.61, samples=30
   iops        : min=  648, max=  678, avg=664.73, stdev= 8.76, samples=30
  lat (usec)   : 10=77.59%, 20=21.88%, 50=0.54%
  fsync/fdatasync/sync_file_range:
    sync (usec): min=449, max=11120, avg=1493.54, stdev=826.17
    sync percentiles (usec):
     |  1.00th=[  469],  5.00th=[  537], 10.00th=[  603], 20.00th=[  701],
     | 30.00th=[  906], 40.00th=[ 1287], 50.00th=[ 1418], 60.00th=[ 1582],
     | 70.00th=[ 1811], 80.00th=[ 2057], 90.00th=[ 2376], 95.00th=[ 2835],
     | 99.00th=[ 4228], 99.50th=[ 4752], 99.90th=[ 6521], 99.95th=[ 7504],
     | 99.99th=[ 9241]
  cpu          : usr=0.28%, sys=2.87%, ctx=30862, majf=0, minf=33
  IO depths    : 1=200.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=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=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=0,10029,0,0 short=10029,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=1

Run status group 0 (all jobs):
  WRITE: bw=1493KiB/s (1529kB/s), 1493KiB/s-1493KiB/s (1529kB/s-1529kB/s), io=21.0MiB (23.1MB), run=15089-15089msec

Disk stats (read/write):
    dm-13: ios=0/26986, merge=0/0, ticks=0/15475, in_queue=15483, util=96.51%, aggrios=0/33352, aggrmerge=0/0, aggrticks=0/14546, aggrin_queue=14552, aggrutil=89.69%
    dm-11: ios=0/33352, merge=0/0, ticks=0/14546, in_queue=14552, util=89.69%, aggrios=2/19669, aggrmerge=0/0, aggrticks=0/7664, aggrin_queue=7666, aggrutil=89.48%
    dm-9: ios=4/5987, merge=0/0, ticks=0/821, in_queue=821, util=4.98%, aggrios=4/39339, aggrmerge=0/0, aggrticks=0/15288, aggrin_queue=15295, aggrutil=94.22%
    dm-8: ios=4/39339, merge=0/0, ticks=0/15288, in_queue=15295, util=94.22%, aggrios=4149/82184, aggrmerge=188/1942, aggrticks=536/8280, aggrin_queue=4864, aggrutil=28.77%
  nvme0n1: ios=4149/82184, merge=188/1942, ticks=536/8280, in_queue=4864, util=28.77%
  dm-10: ios=0/33352, merge=0/0, ticks=0/14507, in_queue=14511, util=89.48%

CentOS 7 - ssd

This is a physical CentOS 7.8 host with a standard consumer SATA SSD device.

mytest: (g=0): rw=write, bs=(R) 2300B-2300B, (W) 2300B-2300B, (T) 2300B-2300B, ioengine=sync, iodepth=1
fio-3.7
Starting 1 process
mytest: Laying out IO file (1 file / 22MiB)
Jobs: 1 (f=1): [W(1)][100.0%][r=0KiB/s,w=957KiB/s][r=0,w=426 IOPS][eta 00m:00s]
mytest: (groupid=0, jobs=1): err= 0: pid=12441: Thu Oct 29 14:49:04 2020
  write: IOPS=428, BW=963KiB/s (986kB/s)(21.0MiB/23389msec)
    clat (usec): min=3, max=107, avg=10.05, stdev= 4.94
     lat (usec): min=3, max=107, avg=10.27, stdev= 4.99
    clat percentiles (nsec):
     |  1.00th=[ 4960],  5.00th=[ 5664], 10.00th=[ 6048], 20.00th=[ 6688],
     | 30.00th=[ 7392], 40.00th=[ 8160], 50.00th=[ 8896], 60.00th=[ 9536],
     | 70.00th=[10432], 80.00th=[11968], 90.00th=[15808], 95.00th=[19584],
     | 99.00th=[32640], 99.50th=[34560], 99.90th=[40704], 99.95th=[44288],
     | 99.99th=[77312]
   bw (  KiB/s): min=  884, max= 1015, per=99.90%, avg=962.02, stdev=41.92, samples=46
   iops        : min=  394, max=  452, avg=428.57, stdev=18.63, samples=46
  lat (usec)   : 4=0.04%, 10=65.40%, 20=30.09%, 50=4.43%, 100=0.03%
  lat (usec)   : 250=0.01%
  fsync/fdatasync/sync_file_range:
    sync (usec): min=764, max=17947, avg=2319.46, stdev=1295.89
    sync percentiles (usec):
     |  1.00th=[  807],  5.00th=[  840], 10.00th=[  848], 20.00th=[  881],
     | 30.00th=[ 1713], 40.00th=[ 1762], 50.00th=[ 1958], 60.00th=[ 2057],
     | 70.00th=[ 2212], 80.00th=[ 4080], 90.00th=[ 4178], 95.00th=[ 4293],
     | 99.00th=[ 4752], 99.50th=[ 6128], 99.90th=[ 6456], 99.95th=[ 6915],
     | 99.99th=[17957]
  cpu          : usr=0.27%, sys=1.94%, ctx=31249, majf=0, minf=32
  IO depths    : 1=200.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=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=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=0,10029,0,0 short=10029,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=1

Run status group 0 (all jobs):
  WRITE: bw=963KiB/s (986kB/s), 963KiB/s-963KiB/s (986kB/s-986kB/s), io=21.0MiB (23.1MB), run=23389-23389msec

Disk stats (read/write):
    dm-0: ios=30/25659, merge=0/0, ticks=34/23133, in_queue=23167, util=97.13%, aggrios=30/31702, aggrmerge=0/146, aggrticks=34/23548, aggrin_queue=23577, aggrutil=97.11%
  sda: ios=30/31702, merge=0/146, ticks=34/23548, in_queue=23577, util=97.11%

RHEL 8.2 - hdd

A RHEL 8.2 host with a 5.4k RPM 2.5" HDD.

fio --rw=write --ioengine=sync --fdatasync=1 --directory=test-data --size=22m --bs=2300 --name=mytest
mytest: (g=0): rw=write, bs=(R) 2300B-2300B, (W) 2300B-2300B, (T) 2300B-2300B, ioengine=sync, iodepth=1
fio-3.7
Starting 1 process
mytest: Laying out IO file (1 file / 22MiB)
Jobs: 1 (f=1): [W(1)][100.0%][r=0KiB/s,w=96KiB/s][r=0,w=43 IOPS][eta 00m:00s]
mytest: (groupid=0, jobs=1): err= 0: pid=26682: Thu Oct 29 14:54:02 2020
  write: IOPS=42, BW=96.4KiB/s (98.8kB/s)(21.0MiB/233578msec)
    clat (usec): min=6, max=101, avg=21.77, stdev= 5.45
     lat (usec): min=7, max=102, avg=22.35, stdev= 5.50
    clat percentiles (nsec):
     |  1.00th=[11584],  5.00th=[14016], 10.00th=[15936], 20.00th=[17536],
     | 30.00th=[18304], 40.00th=[19584], 50.00th=[21120], 60.00th=[22912],
     | 70.00th=[24448], 80.00th=[25984], 90.00th=[28544], 95.00th=[30592],
     | 99.00th=[36608], 99.50th=[39168], 99.90th=[45312], 99.95th=[61696],
     | 99.99th=[99840]
   bw (  KiB/s): min=   58, max=  130, per=99.97%, avg=95.97, stdev= 6.75, samples=467
   iops        : min=   26, max=   58, avg=42.91, stdev= 3.03, samples=467
  lat (usec)   : 10=0.13%, 20=43.28%, 50=56.51%, 100=0.07%, 250=0.01%
  fsync/fdatasync/sync_file_range:
    sync (msec): min=5, max=234, avg=23.26, stdev= 7.68
    sync percentiles (msec):
     |  1.00th=[   11],  5.00th=[   12], 10.00th=[   14], 20.00th=[   16],
     | 30.00th=[   19], 40.00th=[   21], 50.00th=[   24], 60.00th=[   26],
     | 70.00th=[   29], 80.00th=[   31], 90.00th=[   34], 95.00th=[   34],
     | 99.00th=[   34], 99.50th=[   35], 99.90th=[   58], 99.95th=[   64],
     | 99.99th=[  120]
  cpu          : usr=0.04%, sys=0.25%, ctx=29779, majf=0, minf=12
  IO depths    : 1=200.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=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=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=0,10029,0,0 short=10029,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=1

Run status group 0 (all jobs):
  WRITE: bw=96.4KiB/s (98.8kB/s), 96.4KiB/s-96.4KiB/s (98.8kB/s-98.8kB/s), io=21.0MiB (23.1MB), run=233578-233578msec

Disk stats (read/write):
    dm-0: ios=0/31395, merge=0/0, ticks=0/232595, in_queue=232595, util=8.70%, aggrios=0/25792, aggrmerge=0/5628, aggrticks=0/231779, aggrin_queue=219969, aggrutil=8.70%
  sda: ios=0/25792, merge=0/5628, ticks=0/231779, in_queue=219969, util=8.70%

RHEL 8 on vSphere 7 w/VSAN

This test comes from a RHEL 8.1 virtual machine hosted in a vSphere 7 environment using VSAN storage. I do not know the details of the VSAN configuration.

[root@rhel8 ~]# fio --rw=write --ioengine=sync --fdatasync=1 --directory=test-data --size=22m --bs=2300 --name=mytest
mytest: (g=0): rw=write, bs=(R) 2300B-2300B, (W) 2300B-2300B, (T) 2300B-2300B, ioengine=sync, iodepth=1
fio-3.7
Starting 1 process
mytest: Laying out IO file (1 file / 22MiB)
Jobs: 1 (f=1): [W(1)][100.0%][r=0KiB/s,w=375KiB/s][r=0,w=167 IOPS][eta 00m:00s]
mytest: (groupid=0, jobs=1): err= 0: pid=26788: Thu Oct 29 15:51:00 2020
  write: IOPS=182, BW=409KiB/s (419kB/s)(21.0MiB/55073msec)
    clat (usec): min=6, max=12958, avg=28.87, stdev=234.17
     lat (usec): min=6, max=12959, avg=29.25, stdev=234.18
    clat percentiles (usec):
     |  1.00th=[    9],  5.00th=[   10], 10.00th=[   10], 20.00th=[   12],
     | 30.00th=[   13], 40.00th=[   14], 50.00th=[   15], 60.00th=[   16],
     | 70.00th=[   18], 80.00th=[   21], 90.00th=[   60], 95.00th=[   75],
     | 99.00th=[  108], 99.50th=[  123], 99.90th=[  783], 99.95th=[ 5866],
     | 99.99th=[10421]
   bw (  KiB/s): min=   53, max=  759, per=99.80%, avg=408.16, stdev=110.25, samples=110
   iops        : min=   24, max=  338, avg=181.96, stdev=49.09, samples=110
  lat (usec)   : 10=11.96%, 20=65.98%, 50=9.79%, 100=10.89%, 250=1.26%
  lat (usec)   : 500=0.02%, 1000=0.02%
  lat (msec)   : 4=0.01%, 10=0.06%, 20=0.02%
  fsync/fdatasync/sync_file_range:
    sync (usec): min=1198, max=233077, avg=5458.51, stdev=4966.79
    sync percentiles (usec):
     |  1.00th=[  1434],  5.00th=[  1614], 10.00th=[  2040], 20.00th=[  2835],
     | 30.00th=[  3228], 40.00th=[  3621], 50.00th=[  4424], 60.00th=[  6128],
     | 70.00th=[  6652], 80.00th=[  7635], 90.00th=[  8717], 95.00th=[ 10683],
     | 99.00th=[ 17957], 99.50th=[ 22676], 99.90th=[ 47449], 99.95th=[ 70779],
     | 99.99th=[183501]
  cpu          : usr=0.19%, sys=1.53%, ctx=27135, majf=0, minf=13
  IO depths    : 1=200.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=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=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=0,10029,0,0 short=10029,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=1

Run status group 0 (all jobs):
  WRITE: bw=409KiB/s (419kB/s), 409KiB/s-409KiB/s (419kB/s-419kB/s), io=21.0MiB (23.1MB), run=55073-55073msec

Disk stats (read/write):
    dm-0: ios=3/21826, merge=0/0, ticks=3/105013, in_queue=105016, util=39.94%, aggrios=3/21560, aggrmerge=0/364, aggrticks=3/84669, aggrin_queue=76510, aggrutil=40.04%
  sda: ios=3/21560, merge=0/364, ticks=3/84669, in_queue=76510, util=40.04%

Tests from OCP clusters

PortWorx has published an fio container image and sample Kubernetes deployment to test performance. Using this as the basis, I modified the Deployment and jobfile for this test. Anyone can replicate the test using the following:

cat << EOF | oc create -f -
apiVersion: apps/v1
kind: Deployment
metadata:
  name: fio-test
spec:
  strategy:
    rollingUpdate:
      maxSurge: 1
      maxUnavailable: 1
    type: RollingUpdate
  replicas: 1
  selector:
    matchLabels:
      app: fio-test
  template:
    metadata:
      labels:
        app: fio-test
    spec:
      containers:
      - name: fio-container
        image: wallnerryan/fiotools-aio
        ports:
        - containerPort: 8000
        env:
          - name: REMOTEFILES
            value: "https://gist.githubusercontent.com/acsulli/2c7c71594c16273a2cf087963c339568/raw/fd7d07f3dac3e4923a3c08c6d60b03b0e0b63c65/etcd.fio"
          - name: JOBFILES
            value: etcd.fio
          - name: PLOTNAME
            value: etcdtest
---
apiVersion: v1
kind: Service
metadata:
  name: fiotools-etcd
  labels:
    name: fiotools-etcd
spec:
  type: NodePort
  ports:
    - port: 8000
      targetPort: 8000
      name: http
  selector:
    app: fio-test
---
apiVersion: route.openshift.io/v1
kind: Route
metadata:
  name: fiotools-etcd
spec:
  port:
    targetPort: http
  to:
    kind: Service
    name: fiotools-etcd
    weight: 100
EOF

Once the pod was deployed and the fio test finished, browsing to the route provides a directory listing with several files, including a file with the fio output. The contents of that file are shown below.

Azure (OCP 4.6)

etcdtest: (g=0): rw=write, bs=(R) 2300B-2300B, (W) 2300B-2300B, (T) 2300B-2300B, ioengine=sync, iodepth=1
fio-3.8
Starting 1 process
etcdtest: Laying out IO file (1 file / 22MiB)

etcdtest: (groupid=0, jobs=1): err= 0: pid=12: Thu Oct 29 21:27:27 2020
  write: IOPS=359, BW=808KiB/s (828kB/s)(21.0MiB/27873msec)
    clat (usec): min=10, max=99670, avg=31.94, stdev=997.29
     lat (usec): min=11, max=99672, avg=33.22, stdev=997.31
    clat percentiles (usec):
     |  1.00th=[   14],  5.00th=[   15], 10.00th=[   16], 20.00th=[   16],
     | 30.00th=[   17], 40.00th=[   19], 50.00th=[   19], 60.00th=[   20],
     | 70.00th=[   21], 80.00th=[   22], 90.00th=[   31], 95.00th=[   38],
     | 99.00th=[   62], 99.50th=[   81], 99.90th=[  265], 99.95th=[  478],
     | 99.99th=[ 6325]
   bw (  KiB/s): min=   23, max=227722, per=100.00%, avg=120977.34, stdev=28613.46, samples=10029
   iops        : min=    1, max=    1, avg= 1.00, stdev= 0.00, samples=10029
  lat (usec)   : 20=66.86%, 50=31.33%, 100=1.49%, 250=0.21%, 500=0.07%
  lat (usec)   : 750=0.01%, 1000=0.01%
  lat (msec)   : 2=0.01%, 10=0.01%, 100=0.01%
  fsync/fdatasync/sync_file_range:
    sync (usec): min=1023, max=1952.7k, avg=2731.70, stdev=28999.19
    sync percentiles (usec):
     |  1.00th=[   1139],  5.00th=[   1188], 10.00th=[   1221],
     | 20.00th=[   1270], 30.00th=[   1336], 40.00th=[   1483],
     | 50.00th=[   2474], 60.00th=[   2573], 70.00th=[   2638],
     | 80.00th=[   2704], 90.00th=[   2868], 95.00th=[   3163],
     | 99.00th=[   5997], 99.50th=[   8094], 99.90th=[  17695],
     | 99.95th=[  89654], 99.99th=[1887437]
  cpu          : usr=0.58%, sys=3.07%, ctx=29919, majf=0, minf=265
  IO depths    : 1=200.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=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=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=0,10029,0,0 short=10029,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=1

Run status group 0 (all jobs):
  WRITE: bw=808KiB/s (828kB/s), 808KiB/s-808KiB/s (828kB/s-828kB/s), io=21.0MiB (23.1MB), run=27873-27873msec

AWS (OCP 4.6)

etcdtest: (g=0): rw=write, bs=(R) 2300B-2300B, (W) 2300B-2300B, (T) 2300B-2300B, ioengine=sync, iodepth=1
fio-3.8
Starting 1 process
etcdtest: Laying out IO file (1 file / 22MiB)

etcdtest: (groupid=0, jobs=1): err= 0: pid=14: Thu Oct 29 21:33:14 2020
  write: IOPS=1036, BW=2328KiB/s (2384kB/s)(21.0MiB/9676msec)
    clat (nsec): min=3008, max=93601, avg=8060.72, stdev=3770.63
     lat (nsec): min=3097, max=93910, avg=8257.60, stdev=3851.54
    clat percentiles (nsec):
     |  1.00th=[ 4384],  5.00th=[ 4640], 10.00th=[ 4832], 20.00th=[ 5536],
     | 30.00th=[ 6112], 40.00th=[ 6432], 50.00th=[ 6880], 60.00th=[ 7456],
     | 70.00th=[ 8512], 80.00th=[ 9920], 90.00th=[12480], 95.00th=[15040],
     | 99.00th=[22144], 99.50th=[25216], 99.90th=[34048], 99.95th=[36608],
     | 99.99th=[44800]
   bw (  KiB/s): min=24572, max=764627, per=100.00%, avg=326886.33, stdev=103829.66, samples=10029
   iops        : min=    1, max=    1, avg= 1.00, stdev= 0.00, samples=10029
  lat (usec)   : 4=0.23%, 10=80.10%, 20=18.11%, 50=1.56%, 100=0.01%
  fsync/fdatasync/sync_file_range:
    sync (usec): min=356, max=44621, avg=954.29, stdev=1533.39
    sync percentiles (usec):
     |  1.00th=[  388],  5.00th=[  404], 10.00th=[  416], 20.00th=[  433],
     | 30.00th=[  461], 40.00th=[  553], 50.00th=[  922], 60.00th=[  971],
     | 70.00th=[  996], 80.00th=[ 1045], 90.00th=[ 1139], 95.00th=[ 1336],
     | 99.00th=[ 6325], 99.50th=[ 9765], 99.90th=[21890], 99.95th=[29230],
     | 99.99th=[40109]
  cpu          : usr=0.38%, sys=2.18%, ctx=26819, majf=0, minf=266
  IO depths    : 1=200.0%, 2=0.0%, 4=0.0%, 8=0.0%, 16=0.0%, 32=0.0%, >=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=0.0%, 64=0.0%, >=64=0.0%
     issued rwts: total=0,10029,0,0 short=10029,0,0,0 dropped=0,0,0,0
     latency   : target=0, window=0, percentile=100.00%, depth=1

Run status group 0 (all jobs):
  WRITE: bw=2328KiB/s (2384kB/s), 2328KiB/s-2328KiB/s (2384kB/s-2384kB/s), io=21.0MiB (23.1MB), run=9676-9676msec

You have a little typo in the route.opentshift.io metadata.name. 2 leading spaces...

Thank you, I fixed that error and removed the empty labels.

And for OpenShift you need a special right to write on /tmp

We aren't writing to /tmp on the host. It'll write to the overlay file system, which is under /var/lib/containers. Yes, it could affect other Pods that are running or be affected by other running Pods since they are all sharing the same device.

Without the control plane nodes being schedulable, the Pod will not be on a control plane node and as a result the result could be different on a worker. That's ok though since, at least with the default IPI deployment, they're using the same underlying storage type so it should be very similar. FWIW, etcd data is mounted from /var/lib/etcd on the control plane nodes, so if they were schedulable it would be from the same disk, even if you were to mount a different device for /var.

This isn't a perfect test, but that's ok...it's meant to be an estimate for comparison purposes, not a precise performance comparison.

Andrew,
I needed on my 4.5 cluster:
$ oc create serviceaccount useroot
$ oc adm policy add-scc-to-user anyuid -z useroot
$ oc patch deployment/fio-test --patch '{"spec":{"template":{"spec":{"serviceAccountName": "useroot"}}}}'