Score:0

Disk writes suddenly very slow. SMART tests all report normal

cn flag

I noticed that my Ubuntu workstation has become extremely slow when writing to disk. I have two hard drives, one NVME and the other SSD (both set up with ZFS). They should be able to sustain multi-GB/s write speeds; however, benchmarking them shows that they are simply not performing as intended. Running fio --name=random-write --ioengine=posixaio --rw=randwrite --bs=4k --numjobs=1 --size=4g --iodepth=1 --runtime=60 --end_fsync=1 on the NVME gives me:

  WRITE: bw=13.6MiB/s (14.3MB/s), 13.6MiB/s-13.6MiB/s (14.3MB/s-14.3MB/s), io=874MiB (916MB), run=64254-64254msec                                           

Running the same command on the SSD gives me:

  WRITE: bw=45.8MiB/s (48.0MB/s), 45.8MiB/s-45.8MiB/s (48.0MB/s-48.0MB/s), io=2753MiB (2886MB), run=60078-60078msec

Running the same command on a share mounted from my NAS (on a 1GB ethernet connection) gives me:

  WRITE: bw=100MiB/s (105MB/s), 100MiB/s-100MiB/s (105MB/s-105MB/s), io=4096MiB (4295MB), run=40818-40818msec

Yikes! This is pretty bad. How can I identify the problem and repair the issue? I doubt it's a problem specific to the hard drive because I'm seeing this issue on both hard drives.

cc flag
Well, the SSD is 3x as fast as the hdd. If the hdd uses Shingled Magnetic Recording (SMR), that speed is normal (might have to dig a bit, vendor may not be forthcoming on SMR). Search this site for "slow copy" 48MB/s is right in the ballpark for copies which fill up the system buffers.
cn flag
It's not a SMR or CMR HDD. It's a NVMe SSD (Adata SX8200). It's getting only 14.3 MB/s compared to my SATA SSD which is getting 48.0 MB/s.
mangohost

Post an answer

Most people don’t grasp that asking a lot of questions unlocks learning and improves interpersonal bonding. In Alison’s studies, for example, though people could accurately recall how many questions had been asked in their conversations, they didn’t intuit the link between questions and liking. Across four studies, in which participants were engaged in conversations themselves or read transcripts of others’ conversations, people tended not to realize that question asking would influence—or had influenced—the level of amity between the conversationalists.