Raid10 HFS+ perfomance much faster then ZFS

[Roman Kunz]

I was playing around on Maverick with various FS setups. 4x 2TB drives in raidz, raid10 (hfs+/zfs) and it seems raid10 hfs+ outperformes all other setup by at least 80%.
Zfs comp disabled, 128k blocks, only large data gets moved. Arc cache doesn't really kick in as most is write once / read once. Am I missing any tuneables?

I don't really want to use hfs+ again but having on a simple stripe test 2x 2TB read/write with zfs ~120MB/150MB i get with hfs ~200MB/250MB.

-Roman

[Jason Belec]

Well ZFS is about data integrity, everything else second. That said, you seem to be mixing and matching to suit your desired results. Two disks striped with ZFS do not equal raid10 HFS+ in any way reasonable for testing. Perhaps a proper, scientific test, and listing the settings on the pool and file system along with the hardware like drives and controllers. At least then people can compare real world data rather than suggestive statements.

Jason
Sent from my iPhone

> --
>
> ---
> You received this message because you are subscribed to the Google Groups "zfs-macos" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to zfs-macos+...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.

[roman kunz]

You missunderstood. I ran the 2nd raid0 test only to see if there's an overall performance difference. And made the same discovery as with raid10 (and also raid1 btw) that read/write performance on zfs is way below hfs+.

> You received this message because you are subscribed to a topic in the Google Groups "zfs-macos" group.
> To unsubscribe from this topic, visit https://groups.google.com/d/topic/zfs-macos/DKEohkcBCr0/unsubscribe.
> To unsubscribe from this group and all its topics, send an email to zfs-macos+...@googlegroups.com.

[Alex Blewitt]

On 28 Oct 2013, at 09:01, roman kunz <roman...@gmail.com> wrote:
>
> read/write performance on zfs is way below hfs+.

Data security is way above hfs+ too.

Alex

[X Bytor]

On Mon, Oct 28, 2013 at 3:47 AM, Alex Blewitt <alex.b...@gmail.com> wrote:

Data security is way above hfs+ too.

Alex

This. And only this.  I use zfs because my data will not be compromised by one shotgun blast or two gamma ray/alpha particle events at the same time (theoretically). I know I should probably use copies and swap my NAS enclosure off-site once a week, but I have 99.9%  of the failure use-cases covered. If I really need speed, I see a pack of SSD drives in my future...

[Jason Belec]

Yes. Build a small array for speed, but for heavens sake, keep your data on ZFS. I have several clients now who learned the hard way, dead/corrupt controllers, failed raid cards, corrupt data, failed drives, all leading to massive data losses in industries where such things can be lethal. Never mind what happens to you for losing the family photos and videos! ;)

--

Jason Belec

Sent from my iPad

--

[Travis Pavek]

I see similar results with my 4x 1TB raid10.  If you are doing sync writes you will see substantial degrade in write performance.  This is because all data needs to be flushed to the intent log (ZIL) which by default will be on your ZFS pool.  So in effect you are writing your data to disk twice.  While I don't know for sure, I suspect HFS+ is simply throwing the data down wherever there is free space for sync writes.  ZFS is more intelligent on it's writes which is why it has the ZIL.  To test if the ZIL is holding you back you can disable sync writes, making everything async (ie no ZIL).  If it is you can add a small and fast SSD log device where the ZIL will be on the SSD and not your pool.  Even with this option it will still be slower than no ZIL at all.  Of course you should never run production with sync disabled, afterall we are using ZFS because we care about data integrity over performance!

[Daniel Becker]

On Dec 9, 2013, at 9:58 AM, Travis Pavek <tra...@travispavek.com> wrote:
> If you are doing sync writes you will see substantial degrade in write performance.

That's a pretty big if, though. In normal use, the vast majority of writes (including file copies and similar bulk transfers) are generally not synchronous.

[Daniel Jozsef]

I don't remember anyone ever saying that ZFS was fast. In fact it's a resource hog, and quite cumbersome.

It is flexible and safe.

Your test is equivalent to saying that it's much easier to walk around with money in a paper bag than in a reinforced steel security carrier case.

(BTW, speaking of integrity, raid10 is nothing to write home about. It was a surprise to me as well, but no raid configuration protects against silent data corruption. The redundant raid configs were designed to protect from drive failure.)

[Daniel Bethe]

Roman, did you create your zpool with ashift=12 on a 4k drive?  You must manually ascertain whether your drives are 4k because they'll usually lie and be really slow.

--