NVMe Performance Benchmark

We use a hyper-converged NVMe cluster for our Linux V-Server solutions. Below you can see the possible peak values for sequential and random read and write. All tests are carried out for 30 seconds and the resulting average value is used as the result.


Overview

Read Write
Sequential 154 GB/s 35.2 GB/s
Random (IOPS) 800,000 393,000




Sequential reading

fio --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=seq_read.fio --bs=1M --iodepth=64 --size=1G --readwrite=read --numjobs=64 --refill_buffers --time_based --runtime=30



Sequential writing

fio --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=seq_write.fio --bs=1M --iodepth=64 --size=1G --readwrite=write --numjobs=64 --refill_buffers --time_based --runtime=30



Random read (IOPS)

fio --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=seq_read.fio --bs=4K --iodepth=64 --size=1G --readwrite=read --numjobs=8 --group_reporting --refill_buffers --time_based --runtime=30



Random write (IOPS)

fio --ioengine=libaio --direct=1 --gtod_reduce=1 --name=test --filename=seq_write.fio --bs=4K --iodepth=64 --size=1G --readwrite=write --numjobs=8 --group_reporting --refill_buffers --time_based --runtime=30