FreeNAS/ZFS woes (was Re: Is bcache ready for enterprise production?)

Tom Buskey tom at buskey.name
Mon Aug 18 09:29:21 EDT 2014


Thanks for the detailed posts on FreeNAS/ZFS/FreeBSD!

I agree with you on deduplication.  Just know unless you have infinite
RAM.  I found out at home, running Solaris, with ~ 2TB on a 3 GB system.  I
had to put the disks on a system with 8 GB or RAM and it took 3-4 days to
import the zpool so I could read it.  I copied it elsewhere as quickly as
possible.  I avoid dedupe.

Solaris has a very efficient virtual memory system and uses it
extensively.  Linux and FreeBSD are different.   I think the memory tuning
for ZFS is still not there yet.

I've used ZFS at work with its integrated NFS.  It worked well with
Solaris, Linux and Windows clients.  I did not use the SMB function.

I've also used the snapshotting.  It worked very well with ClearCase vobs
and views on top of it.

I'm currently using ZFS on Linux at home.  I won't use it at work.

  I find the "quotas" or dynamic "partition" resizing to be way easier then
using LVM.  LVM needs a umount, ZFS doesn't.  And snapshotting is nice to
have.  If that's all I use, I basically have the NetApp of 2000 on gigabit
ethernet.  No SAN features, just NFS and it's very functional with low
maintenance.  And I get the benefit of error correction and self healing if
I do RAID.

I have not done anything with iSCSI.  Solaris has a standard, tested,
tuned, integrated iSCSI with ZFS.

Linux is a little bit less settled.  It's not integrated with ZFS.  NFS
exports are not integrated, so I use /etc/exports.  I don't think SMB is
integrated in Linux and I use samba v3.

I've heard about issues deleting zvols taking a long time in Solaris
(weeks), but not crashing.

What you did on FreeNAS shouldn't be out of the ordinary.  I'd be a little
curious to see if using FFS instead of ZFS would be more stable.  Or
current FreeBSD + OpenZFS would be better.  But not enough to test it.

I'd also like to see more on btrfs in real use.  It will get more attention
on Linux for stability, RAM use, etc then ZFS ever will.  Until btrfs is
ready, I'll stick with ext3/4 and XFS at work on Linux.

Is anyone using btrfs?






On Sat, Aug 16, 2014 at 8:27 PM, Alan Johnson <alan at datdec.com> wrote:

> On Fri, Aug 15, 2014 at 11:35 AM, Tom Buskey <tom at buskey.name> wrote:
>
>> I've never used FreeNAS, but I've used ZFS quite a bit.
>>
>> I've used it natively on Solaris at work and on Ubuntu with ZFS on Linux
>> at home.
>>
>> I've been very happy with it.  There is information out there on
>> tuning/not tuning on Solaris.  ZFS on Linux (and BSD) has different tunings.
>>
>> From the short skim I just did on bcache and its support in CentOS 7, it
>> sounds like it's much less evolved then ZFS on Linux or FreeNAS.
>>
>
> bcache is certainly newer than ZFS.  It started in part as a response to
> the caching features ZFS offers, IIRC.  It is also arbitrarily simpler than
> ZFS, meant to be a bolt on to any block device storage management you are
> already doing, not a full replacement.  In my case, it would be a software
> 36TB RAID6 as the backing device with one hot spare, a 500GB SSD mirror for
> the cache device, with LVM and mostly ext4 on top of that.
>
> Googling around for alternatives, I came across
> *Why We Recommend Against OpenFiler
> <http://community.spiceworks.com/topic/373443-why-we-recommend-against-openfiler>
> ​.* I don't know the author from Adam, but he come across as very
> knowledgeable to me and he says this in response to one of the comments:
>
>
> The biggest advantage of FreeNAS is that it offers ZFS which can be very
> cool but ZFS has so much [hype] around it that it tends to cause bad
> decision making, a [belief] that parity RAID has been magically fixed [e.g]
> and people forget that the biggest hype is current ZFS which is Solaris
> only, [FreeNAS] is an older version.
>
>
> That was from a year ago though.  I don't know if it holds true in FreeNAS
> 9.2, which I had upgraded to long before my box blew up.​  Regardless, the
> Solaris experience is likely to be very different.  I would be curious if
> you could setup a test to see how ZFS on Solaris handles deleting zvols
> from a deduplicated zpool.
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.gnhlug.org/mailman/private/gnhlug-discuss/attachments/20140818/b5feff45/attachment.html 


More information about the gnhlug-discuss mailing list