3

I'm trying to plan my storage for the home usage on the Linux machine. I have one 250 GB SSD for OS, plus 4 TB HDD for storage, which I'm going to use as zpool.

According to this thread: performance declines at about 85% utilization.

Math is simple 4000GB x 0.15 = 600GB.

Hmm, 600GB for a home user is a lot of wasted space.

I wonder if I could partially solve this problem by adding the dedicated Cache / Log partitions on my SSD? Or is it 85% is 85%, no matter what business?

2
4

Please heed the warning.

Don't plan a new solution and expect to use 80% or more of the storage space immediately. If that means you need to buy more disks, buy more disks!

4
  • Thank you for your answer, but it still leaves me in doubt whether adding ZIL would allow me to utilise the disk to the higher percentage without noticeable performance penalty? If I've got it right, with ZIL ZFS should first try writing to it, then lazily offload the data to the other devices in the pool. My zpool will be used mainly for archival purposes, and I expect burst of ~20GB at once, hence no frequent writing, no growing logs. But I still would like that my HDD was speedier than a USB stick when I have ~0.5 TB of free space... – NarūnasK Jul 30 '16 at 23:57
  • 2
    ZIL only helps with synchronous writes to zpool. Synchronous writes are writes that are guaranteed to be in stable storage when the write operation returns. ZIL device makes these writes faster, since the intent log can be used to finish writes to final media in case of a system crash, and so the faster writing speed of SSD ZIL device makes synchronous writes faster. In most cases, writes are asynchronous, where write operations can be cached in memory, and the write operation can return before data is written to stable media. Neither ZIL or memory cache help with the fragmentation of ZFS. – Tero Kilkanen Jul 31 '16 at 1:37
  • 1
    @TeroKilkanen All ZFS pools have an intent log (ZIL). You are referring to a SLOG device, which puts the ZIL on a device separate from those storing the bulk data for long-term access. – user Aug 17 '16 at 14:06
  • @MichaelKjörling You are right. Although my intention was to point out that the issue is as described in constantin.glez.de/blog/2010/07/… , I failed to explain it correctly. Shame on me. – Tero Kilkanen Aug 17 '16 at 14:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.