From https://cloud.google.com/compute/docs/compute-optimized-machines#h3_limitations
Persistent Disk and Google Cloud Hyperdisk performance is capped at 15,000 IOPS and 240 MBps throughput.
This means the initial fio file write takes about an hour to complete (500GB @ 240MB/s) before the real tests even begin.
With the default timeout of 20m for this, the storageperf tests for h3 instances will time out unless the perfect speed is maintained for a whole hour.
I think a good solution would be write a smaller amount of data for h3 instances, perhaps 20G or 50G instead of 500 for the initial pre-test write.
root 82318 0.0 0.0 1244224 23324 ? Sl 14:45 0:00 \_ /etc/image_test2047922026/image_test -test.v -test.timeout 20m -test.run TestRandomReadIOPS|TestSequentialReadIOPS|TestRandomWriteIOPS|TestSequentialWriteIOPS
root 82516 0.4 0.0 239496 4928 ? Sl 14:46 0:08 \_ fio --name=fill_disk --direct=1 --verify=0 --randrepeat=0 --bs=128K --iodepth=64 --rw=randwrite --iodepth_batch_submit=64 --iodepth_batch_complete_max=64 --ioengine=libaio --filesize=500G --filename=/dev/nvme0n2
root 82686 3.5 0.0 182592 9564 ? Ss 14:46 1:01 \_ fio --name=fill_disk --direct=1 --verify=0 --randrepeat=0 --bs=128K --iodepth=64 --rw=randwrite --iodepth_batch_submit=64 --iodepth_batch_complete_max=64 --ioengine=libaio --filesize=500G --filename=/dev/nvme0n2
[root@h3standard88pdbalanced-storageperf-6j2t8 jonathan]# iostat -x 2
Linux 5.14.0-570.49.1.el9_6.x86_64 (h3standard88pdbalanced-storageperf-6j2t8) 10/01/2025 _x86_64_ (88 CPU)
avg-cpu: %user %nice %system %iowait %steal %idle
0.02 0.00 0.04 0.00 0.00 99.94
Device r/s rkB/s rrqm/s %rrqm r_await rareq-sz w/s wkB/s wrqm/s %wrqm w_await wareq-sz d/s dkB/s drqm/s %drqm d_await dareq-sz f/s f_await aqu-sz %util
nvme0n1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
nvme0n2 0.00 0.00 0.00 0.00 0.00 0.00 1920.00 245760.00 0.00 0.00 33.30 128.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 63.94 100.0
From https://cloud.google.com/compute/docs/compute-optimized-machines#h3_limitations
This means the initial fio file write takes about an hour to complete (500GB @ 240MB/s) before the real tests even begin.
With the default timeout of 20m for this, the storageperf tests for
h3instances will time out unless the perfect speed is maintained for a whole hour.I think a good solution would be write a smaller amount of data for h3 instances, perhaps 20G or 50G instead of 500 for the initial pre-test write.