Very bad write speed on mdadm RAID1 with SSDs

Question

I have two Samsung 860 QVO in RAID1 with mdadm. The write speed of a single disc is around 450MB/s. I reach a maximum of 70MB/s on the RAID 1. The reading speed is normal.

xfs is used as the file system.

What is the problem here?

/proc/mdstat

Personalities : [linear] [multipath] [raid0] [raid1] [raid6] [raid5] [raid4] [raid10]
md0 : active raid1 sdb[0] sdc[1]
      976630464 blocks super 1.2 [2/2] [UU]
      bitmap: 3/8 pages [12KB], 65536KB chunk

unused devices: <none>

dd if=/dev/zero of=/vm-storage/testfile bs=1G count=1 oflag=dsync

1073741824 bytes (1,1 GB, 1,0 GiB) copied, 16,6422 s, 64,5 MB/s

score 1 · Answer 1 · answered Jul 31 '20 at 15:00

1

Your writes are slow because you used oflag=dsync. As the man page states:

       dsync  use synchronized I/O for data

This means that the system waits after every write until the storage has confirmed it is written. Generally you wouldn't use synchronized I/O unless your application desperately needed to ensure that a single write was actually committed to the media before continuing.

answered Jul 31 '20 at 15:00

Michael Hampton

244,070
43
506
972

How do I make a meaningful benchmark? – DerFlo129 Jul 31 '20 at 15:53
@DerFlo129 Don't use `oflag=dsync`. – Michael Hampton Jul 31 '20 at 15:57
Simply no oflag use? – DerFlo129 Jul 31 '20 at 16:14
@DerFlo129 Right, you don't need it. – Michael Hampton Jul 31 '20 at 16:15
1

@MichaelHampton is right. Normal applications won't use `dsync`. For an accepted storage benchmark, try the `fio` application. `dd` is just not built to drive storage subsystems. – Mike Andrews Aug 01 '20 at 05:24

Very bad write speed on mdadm RAID1 with SSDs

1 Answers1