ZFS - no thanks!

2007-08-11 Thread Wojciech Puchar

just ended testing.

after having all my data  (test system fortunately) on ZFS including root, 
i lost /boot partition, which was on pendrive to make testing easier.


well - no problem - i've started my normal 6.2 syste, got bootonly CD, 
removed mfsroot, added (as on ZFS system) 
vfs.root.mountfrom=zfs:tank/root, put tu pendrive, bsdlabel -B etc..


started..

it CANNOT mount zfs.

well - i don't have live CD

the only thing left was 7.0 livecd  but amd64 only.

well - i have qemu.

started qemu with this image, added -hda with my ZFS disk, started.

got to fixid, kldload zfs.ko

zfs import - well all's fine

zfs import -f tank

after some time done

tried any command - looks like no libraries. well - probably it 
overmounted / from zfs...



next reboot

now

zfs import -R /mnt -f tank


and it's waiting forever. doing nothing, both CTRL-C and CTRL-Z doesn't 
work.


it's doing nothing as i see that qemu uses almost no CPU.


EXCELLENT filesystem with excellent protection of my data



on the other hand it's faster over UFS on small files but not that faster.
it uses HUGE amount of RAM.

it's set copies=n is a joke.
you have no warranty where are the copies (often on the same disk).
zpool scrub DOES NOT move copies to other disk from the same when other is 
made available!!

raidz can't be expanded

cache flushing CAN NOT be disable for selected device, only for 
everything.


my USB-IDE converter make doesn't allow it, but my 2.5 does!
i use USB-IDE converted with disk as a backup. with ZFS it's impossible 
unless i will turn off flushing for everything - losing it's important 
adventage.



disks based pool (no mirror/raidz) won't start AT ALL with one element 
unattached!!! EVEN if everything has copies1 !!!



summary: excellent idea turned into pile of s...t

no thanks
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]


ZFS - no thanks! - few more words

2007-08-11 Thread Wojciech Puchar
for those who got much better experience with ZFS and for everyone else 
using anything:


please do remember - no RAID hardware or software, no zfs, no anything is 
a replacement for REGULAR BACKUPS done on removable media or different 
machine. in second case at least sometimes it should be done to removable 
media too.


with this way i never lost any data, not using any mirrors or RAID-5 ever.
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]


Re: ZFS - no thanks!

2007-08-11 Thread Christian Walther
Hi Wojciech,

let me start with pointing out that ZFS is still an experimental feature.
Secondly, this is the wrong list, because ZFS is a feature of FreeBSD-CURRENT.
Adding freebsd-current to CC, maybe one of the ZFS developers can give
their $0.02.

On 11/08/07, Wojciech Puchar [EMAIL PROTECTED] wrote:
 just ended testing.

 after having all my data  (test system fortunately) on ZFS including root,
 i lost /boot partition, which was on pendrive to make testing easier.

 well - no problem - i've started my normal 6.2 syste, got bootonly CD,
 removed mfsroot, added (as on ZFS system)
 vfs.root.mountfrom=zfs:tank/root, put tu pendrive, bsdlabel -B etc..

If I understand you correctly you tried to use ZFS with FreeBSD 6.2. Won't work.

 on the other hand it's faster over UFS on small files but not that faster.
 it uses HUGE amount of RAM.

ZFS probably isn't for Joe average user. No offense ment with this,
but ZFS does has a nice set of features, but not many are needed for
every days work. At least if you're on a workstation.

 it's set copies=n is a joke.
 you have no warranty where are the copies (often on the same disk).

Yes, because copies=n isn't there to protect against an entire disk
failing. It protects data against block failures. So if an individual
block on the disk wents bad, as it's most often the case when a disk
starts to die, you have some more backups with the correct checksum.

 zpool scrub DOES NOT move copies to other disk from the same when other is
 made available!!

No, because it's not dedicated to do this. AFAIK zfs does this
automagically every time you attach a disk and add it to a certain
zpool.
scrubbing is in cases where disk faults have been found. It means that
the ZFS compares the data and it's checksum, thus locating any trouble
and fixes it.

 raidz can't be expanded

 cache flushing CAN NOT be disable for selected device, only for
 everything.

 my USB-IDE converter make doesn't allow it, but my 2.5 does!
 i use USB-IDE converted with disk as a backup. with ZFS it's impossible
 unless i will turn off flushing for everything - losing it's important
 adventage.


 disks based pool (no mirror/raidz) won't start AT ALL with one element
 unattached!!! EVEN if everything has copies1 !!!

I don't know your setup but for me it works fine.
I'm currently at the Chaos Communication Camp, next to me is an
AthlonXP powered with FreeBSD-CURRENT installed. 3x400GB HDD using
ZFS, giving 732GB net capacity. 2GB RAM.
We had a few issues with the hardware. One of the disks is unstable,
leading to crashes eventually. I started the system without the disk,
I changed the disks position, moved them from a PCI Controller to the
onboard controller and stuff.
ZFS came up fine.
I'm really impressed with ZFS and it's features.

The system is pretty busy, we've 15 users max with 1MBit/sec allowed.
Firewall states a throughput of 90MBit/sec upstream, saturating the
100MBit/sec NIC.

With ZFS you're making use of all your HW, CPU, RAM, PCI Bus etc. So
if something 's wrong with your HW you'll notice. But that doesn't
necessarily mean that it's related to ZFS.
For example I encountered a poor system performance, with lots of
interrupts. Tried to tweak ZFS a bit, didn't make a difference. Then I
took my primary slave disk from the PCI Controller, attached it to
onboard primary master - and things went out fine.

Just as a side note: If your dmesg reports something like this:
atapci0: ITE IT8212F UDMA133 controller port
0xdc00-0xdc07,0xd800-0xd803,0xd400-0xd407,0xd000-0xd003,0xcc00-0xcc0f
irq 10 at device 6.0 on pci0
...remove it. ;-)

Christian
___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]


Re: ZFS - no thanks!

2007-08-11 Thread Kris Kennaway
On Sat, Aug 11, 2007 at 01:25:44PM +0200, Wojciech Puchar wrote:
 just ended testing.

Wow, that was quick.  It looks like you made some very hasty
judgements, and a lot of the problems you encountered were quite
frankly your own fault.

 after having all my data  (test system fortunately) on ZFS including root, 
 i lost /boot partition, which was on pendrive to make testing easier.
 
 well - no problem - i've started my normal 6.2 syste, got bootonly CD, 
 removed mfsroot, added (as on ZFS system) 
 vfs.root.mountfrom=zfs:tank/root, put tu pendrive, bsdlabel -B etc..
 
 started..
 
 it CANNOT mount zfs.

Yes, clearly documented.

 well - i don't have live CD

Easy to download one.

 the only thing left was 7.0 livecd  but amd64 only.
 
 well - i have qemu.
 
 started qemu with this image, added -hda with my ZFS disk, started.
 
 got to fixid, kldload zfs.ko
 
 zfs import - well all's fine
 
 zfs import -f tank
 
 after some time done
 
 tried any command - looks like no libraries. well - probably it 
 overmounted / from zfs...

I doubt it.  More likely you are having problems from trying to use
your i386 system on an amd64 kernel, which will be looking in a
different place for the i386 libraries.

 next reboot
 
 now
 
 zfs import -R /mnt -f tank
 
 
 and it's waiting forever. doing nothing, both CTRL-C and CTRL-Z doesn't 
 work.
 
 it's doing nothing as i see that qemu uses almost no CPU.

You may have found a bug, unfortunately you needed to obtain more
debugging information to be a useful report.

 it uses HUGE amount of RAM.

Documented.

 it's set copies=n is a joke.
 you have no warranty where are the copies (often on the same disk).

Not according to the documentation.

 zpool scrub DOES NOT move copies to other disk from the same when other is 
 made available!!

It should, I think.

 raidz can't be expanded

Dunno what you mean by expanded.

 cache flushing CAN NOT be disable for selected device, only for 
 everything.
 
 my USB-IDE converter make doesn't allow it, but my 2.5 does!
 i use USB-IDE converted with disk as a backup. with ZFS it's impossible 
 unless i will turn off flushing for everything - losing it's important 
 adventage.

Not sure what you mean here either.

 disks based pool (no mirror/raidz) won't start AT ALL with one element 
 unattached!!! EVEN if everything has copies1 !!!

That's not what copies is supposed to be used for.  If you want
degraded mode, use a mirror/raid configuration,.

 summary: excellent idea turned into pile of s...t

summary: user had many incorrect expectations about the software and
is not willing to correctly report possible bugs so they can be fixed,
therefore should avoid running all pre-release versions of freebsd.

Kris


pgpUhimTHddG0.pgp
Description: PGP signature


Re: ZFS - no thanks!

2007-08-11 Thread Wojciech Puchar

I doubt it.  More likely you are having problems from trying to use
your i386 system on an amd64 kernel, which will be looking in a
different place for the i386 libraries.


i was using qemu to emulate amd64. tried i386 live CD 5 minutes ago.

SAME EFFECT! system hangs while doing import!

no crash so i can't do a crashdump :(


you said that i did quick testing. yes - because before it i already new 
what i want to test.

___
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to [EMAIL PROTECTED]