zfs discuss - Jul 2010 - Confused about consumer drives and zfs can someone help?

I wanted to build a small back up (maybe also NAS) server using 
OpenSolaris and ZFS using consumer drives but after reading a number of 
threads and blogs I''m totally confused and was hoping I could get some 
questions answered since many people have been using consumer drives 
with zfs.

When ZFS first was announced I remember reading that it''s error 
correction could make consumer level disks suitable for RAID 
applications but that doesn''t seem to be the case. I''m
familiar and have
had good experience with WD drives so I''ll be mainly referring to them.

1.  WD Caviar Black
************************
Can they be used with in raidz or mirrors?

With the new models the firmware is locked and you can''t change the
TLER
settings. When a drive detects an error it''s going to hang a long time 
trying to correct the problem before it reports the error to ZFS right? 
What type of errors are we talking about, bad blocks or something more 
serious?

Once it reports an error zfs marks the drive as offline? In the case of 
something like bad blocks, can the drive somehow be salvaged by marking 
off the bad blocks and reducing the size of the vdev or does it need to 
be replaced?

Because of the TLER issue, does this mean that drives that would have 
been able to remain in service as just an ordinary stand alone drive 
would need to be replaced so that you can expect to replace consumer 
drives like the WD Caviar Blacks to not last as long using ZFS?


2. WD Caviar Greens
*************************
I was hoping to use low power drives like the WD Caviar Greens. In 
addition to the TLER issue they have an issue with too many load/unload 
cycles that would wear out the drive faster when used in raid configs 
including zfs. WD put out a utility that could increase the time to 
reduce the load/unload cycles (wdidle.exe). Does this still work with 
the caviar green drives? Even if it does work, changing the idle 
behavior is going to make it use more energy right?

Are Caviar Green drives (the new ones anyway) completely unsuitable for 
a ZFS based back-up or NAS server?

3. WD Scorpio Blue/Black
*******************************
It seems like the greens are out so these might be a better low power 
option? These are the 2.5" SATA laptop drives. These seem like a great 
drive to use in a small backup server because of their size and power 
usage. If you don''t need large capacity but want multiple drives for 
redundancy could these be used? Storage density isn''t so bad
considering
you can fit 2 in the space of one 3.5" drive.

It would be great if these drives could be used but I haven''t been able
to find much info using them with zfs. To be able to save money and 
space it would be nice to at least have the option of using 2 of these 
mirrored as the rpool.

The 5400RPM Scorpio Blue drives with their low power requirements and 
low heat would make a nice home backup server if they could work with 
ZFS. I''ve read about people using them in a RAID0 config in their 
laptops so maybe there''s hope? There''s a 1TB Scorpio blue that
uses
"Advanced Formatting" (whatever that is) will that be a problem for
ZFS?
4 of them can fit in 1 5.25" bay and cost around $700 for 3TB.

4. Consumer vs High End Controllers
********************************************
I read a something along the lines of ''you can use consumer drives with
low end sata cards but the enterprise drives like the RE3 and 
Velociraptor need to be used with high end sata raid cards''.

Can someone clarify this for me please? I was planning  on using 
something like the Syba SY-PCI40010 
(http://www.newegg.com/Product/Product.aspx?Item=N82E16816124028&cm_re=Sil3124-_-16-124-028-_-Product)
which uses a SiL3124 chipset. I''m not great with this type of hardware 
but my understanding is that even though it has RAID features it''s a 
software raid card and when you don''t use the RAID features
it''s just a
plain SATA port multiplier.

This really confused me because I thought any hardware or software raid 
card is just a sata port replicator when you''re not letting the card do
the RAID.

In terms of performance, problems, reliability, drive life, what can I 
expect under the 4 variations:

SiL Controller with caviar drives?

Sil Controller with RE3 drives?

High end raid controller with caviar drives?

High end raid controller with RE3 drives?

5. Mirror vs raidz
**********************
Can any of the issues with consumer drives be reduced using one type of 
vdev over the other?  Will adding a seperate log or cache device help?

For a backup server, which would you choose, 4 drives in raidz, 4 drives 
in raidz2, 3 drives in raidz with hs, 4 drives with 2 mirrored pairs?

On Wed, Jul 21, 2010 at 1:38 PM, JavaWebDev <javawebdev at verizon.net>
wrote:
> 1. ?WD Caviar Black
> ************************
> Can they be used with in raidz or mirrors?
We use the 500 GB versions attached to 3Ware controllers (configured
as Single Disk arrays).  They work quite nicely.
> With the new models the firmware is locked and you can''t change
the TLER
> settings. When a drive detects an error it''s going to hang a long
time
> trying to correct the problem before it reports the error to ZFS right?
What
> type of errors are we talking about, bad blocks or something more serious?
Haven''t noticed any issues.   The RAID controller notes the odd (maybe
1 per fortnight) timeout, but then reconnects the drive without ZFS
noticing.  We have yet (2 years) to have a drive timeout and cause ZFS
to offline the drive.
> 2. WD Caviar Greens
> *************************
> I was hoping to use low power drives like the WD Caviar Greens. In addition
> to the TLER issue they have an issue with too many load/unload cycles that
> would wear out the drive faster when used in raid configs including zfs. WD
> put out a utility that could increase the time to reduce the load/unload
> cycles (wdidle.exe). Does this still work with the caviar green drives?
Even
> if it does work, changing the idle behavior is going to make it use more
> energy right?
>
> Are Caviar Green drives (the new ones anyway) completely unsuitable for a
> ZFS based back-up or NAS server?
We use 8x 1.5 TB WD Green drives, since they were really inexpensive
at the time, and looked good on paper.  We''ve been regretting it ever
since.  I actively pray for these to die so that I can replace them
with Seagate 1.5 TB drives.

For a home setup where you don''t care about disk throughput, these may
be workable.  Otherwise, avoid the entire Green/GP series of WD disks.
 They are, to put it nicely, crap.  Don''t ask for an elaboration
unless you want a multiple page diatribe on the horrors of using these
drives.  Just don''t use them outside of the home.

We also use WD Caviar Blue 500 GB, and RE2/RE3 500 GB drives attached
to 3Ware controllers without issues.

Finally, we also use Seagate Barracuda 7200.11 500 GB and 1.5 TB
drives with great success.  As wel as Seagate ES.2 500 GB drives.

Basically, anything other than the WD Green/GP drives works well.
> 5. Mirror vs raidz
> **********************
> Can any of the issues with consumer drives be reduced using one type of
vdev
> over the other? ?Will adding a seperate log or cache device help?
An L2ARC will help with reads, especially if using raidz vdevs.
> For a backup server, which would you choose, 4 drives in raidz, 4 drives in
> raidz2, 3 drives in raidz with hs, 4 drives with 2 mirrored pairs?
Only 4 drives?  Do they still make servers that only use 4 drives?  ;)

It all depends on which you prefer, raw throughput or redundancy.

For best performance, use 2x 2-drive mirrors.
For best redundancy, use 1x 4-drive raidz2.
For middle-of-the-road performance/redundancy, use 1x 4-drive raidz1.

Note:  newegg.ca has a sale on right now.  WD Caviar Black 1 TB drives
are only $85 CDN.

-- 
Freddie Cash
fjwcash at gmail.com

> I wanted to build a small back up (maybe also NAS) server using 
A common question that I am trying to get answered (and have a few) here:
http://www.opensolaris.org/jive/thread.jspa?threadID=102368&tstart=0

Rob
-- 
This message posted from opensolaris.org

Using a new email client and didn''t notice that I didn''t reply
to the
list. Since it might be helpful to others here are the missing bits.
>> On 7/21/2010 5:07 PM, Freddie Cash wrote:
>>
>> We use the 500 GB versions attached to 3Ware controllers (configured
>> as Single Disk arrays).  They work quite nicely.
>>
> Thanks, your reply was very helpful.
>
> Did you use the TLER utility to change your Caviar drives? Have you 
> installed any of the new Caviar drives that don''t support changing
> TLER settings?
>
> I forgot to mention, this is for a home server. I want something to 
> backup my other computers and provide some network storage for stuff 
> that doesn''t that doesn''t get used often but gets used
frequently
> enough or are too big that optical storage would be a pain.
>
> Something along the lines of this 
> http://blogs.sun.com/mebius/entry/diy_home_nas_box_with2 or smaller if 
> I could use the 2.5" Scorpio drives. I''m willing to sacrifice
> performance for power and heat savings. This is the only thread I 
> found regarding zfs and scorpio drives 
>
http://mail.opensolaris.org/pipermail/opensolaris-discuss/2009-June/048287.html
>
> It sounds like your 3Ware controller might be catching the errors 
> before it gets to zfs but if zfs had to handle it then it might have 
> the problems others have reported. If that''s the case, *why
can''t zfs
> do what the 3ware controller is doing?* Or does it and these reports 
> about problems with consumer drives and TLER are overblown? A <20s 
> delay every couple of weeks I can live with.
>
> With the mini-itx mobo I''m planning to use I''m limited to
PCI cards
> and there isn''t much selection. The SiL based cards are supported
in
> OpenSolaris but I''m not sure if they''re going to behave
the same as
> your 3ware card.
>
> I found this quote here http://hardforum.com/showthread.php?t=1500505 
> which contradicts a lot of what I read regarding ZFS and consumer drives.
>
> "Do i need to use TLER or RAID edition harddrives?
> No and if you use TLER you should disable it when using ZFS. TLER is 
> only useful for mission-critical servers who cannot afford to be 
> frozen for 10-60 seconds, and to cope with bad quality RAID controller 
> that panic when a drive is not responding for multiple seconds because 
> its performing recovery on some sector. Do not use TLER with ZFS!
>
> Instead, allow the drive to recover its errors. ZFS will wait, the 
> wait time can be configured. You won''t have broken RAID arrays,
which
> is common with Windows-based FakeRAID arrays."
And Freddie Cash replied :
> - hide quoted text -
>> >  On 7/21/2010 5:07 PM, Freddie Cash wrote:
>> >  We use the 500 GB versions attached to 3Ware controllers
(configured
>> >  as Single Disk arrays).  They work quite nicely.
>> >
>> >  Thanks, your reply was very helpful.
>> >
>> >  Did you use the TLER utility to change your Caviar drives? Have
you
>> >  installed any of the new Caviar drives that don''t
support changing TLER
>> >  settings?
> The wdtler utility doesn''t work on the Caviar Green drives.
>
> We did use the wdidle3 utility, though, to "disable" the 8 second
idle timeout.
>
> Haven''t needed to use the wdtler utility on any of the Caviar
Black or
> Caviar Blue drives.  They don''t cause any issues.
>
> - hide quoted text -
>> >  I forgot to mention, this is for a home server. I want something
to backup
>> >  my other computers and provide some network storage for stuff
that doesn''t
>> >  that doesn''t get used often but gets used frequently
enough or are too big
>> >  that optical storage would be a pain.
>> >
>> >  Something along the lines of this
>> >  http://blogs.sun.com/mebius/entry/diy_home_nas_box_with2  or
smaller if I
>> >  could use the 2.5" Scorpio drives. I''m willing to
sacrifice performance for
>> >  power and heat savings. This is the only thread I found regarding
zfs and
>> >  scorpio drives
>> > 
http://mail.opensolaris.org/pipermail/opensolaris-discuss/2009-June/048287.html
> If you want low-power, then the Green drives will be fine.  So long as
> you don''t want super-fast throughput.  These are 5900 RPM drives
that
> can slow down to 3400 RPM or thereabouts.  But you definitely want to
> play with wdidle3 to change the default head-parking idle timeout.
> Otherwise you''ll burn through the 500,000 load-store cycles in a
> couple of months.
>
> - hide quoted text -
>> >  It sounds like your 3Ware controller might be catching the errors
before it
>> >  gets to zfs but if zfs had to handle it then it might have the
problems
>> >  others have reported. If that''s the case, why
can''t zfs do what the 3ware
>> >  controller is doing? Or does it and these reports about problems
with
>> >  consumer drives and TLER are overblown? A<20s delay every
couple of weeks I
>> >  can live with.
> No idea.  I don''t have any ZFS systems using non-RAID controllers
yet.
>
> -- Freddie Cash fjwcash at gmail.com


-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20100722/eb7c4dc5/attachment.html>

There is alot there to reply to...but I will try and help...

Re. TLER. Do not worry about TLER when using ZFS. ZFS will handle it either way
and will NOT time out and drop the drive...it may wait a long time, but it will
not time out and drop the drive - nor will it have an issue if you do enable
TLER-ON (which sets time out to 7 seonds).  I run both with TLER-ON (disks from
an "old" mdadm raid array) and without (TLER-OFF) 1.5TB WDEADS.

I can not speak for the WD EARS, but the WDEADS are fine in my home nas. I also
run 1.5TB Samsung Green/Silencer series and Seagate 11''s. Others swear
by Hitachi. I would recommend the Samsung or Hitachi and not the new WD EARS
which have that 4k sectors or whatever it is.

Re the CPU, do not go low power Atom etc, go a newish Core2 duo...the power
differential at idle is bugger all and when you want to use the nas, ZFS will
make good use of the CPU. Honestly, sit down and do the calculations on power
savings of a low power cpu and you''ll see it''s better to just
not have that 5th beer on a friday - you''ll save more money that way
and be MUCH happier with your nas performance.

re. cards...I use and recommend these 8-Port SUPERMICRO AOC-USASLP-L8I UIO SAS.
They are cheap on e-bay, just work and are fast. Use them.

You do want alot of ram I use 8GB, but you can use 4. Ram is cheap, ZFS loves
ram, just buy 8.

IMHO (and that of the best practice guide) - you should mirror the rpool (o/s
disk). Just buy 2 cheap laptop drives and when installing choose to
"mirror" them.

I hope that helps.
-- 
This message posted from opensolaris.org

I''ve found the Seagate 7200.12 1tb drives and Hitachi 7k2000 2TB drives
to
be by far the best.

I''ve read lots of horror stories about any WD drive with 4k
sectors....it''sbest to stay away from them.

I''ve also read plenty of people say that the green drives are terrible.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://mail.opensolaris.org/pipermail/zfs-discuss/attachments/20100723/444ddce2/attachment.html>

On 7/23/2010 3:39 AM, tomwaters wrote:
> There is alot there to reply to...but I will try and help...
>
> Re. TLER. Do not worry about TLER when using ZFS. ZFS will handle it either
way and will NOT time out and drop the drive...it may wait a long time, but it
will not time out and drop the drive - nor will it have an issue if you do
enable TLER-ON (which sets time out to 7 seonds).  I run both with TLER-ON
(disks from an "old" mdadm raid array) and without (TLER-OFF) 1.5TB
WDEADS.
That''s reassuring. I guess the only time the issue will come up is when
the drive starts to develop errors.
> I can not speak for the WD EARS, but the WDEADS are fine in my home nas. I
also run 1.5TB Samsung Green/Silencer series and Seagate 11''s. Others
swear by Hitachi. I would recommend the Samsung or Hitachi and not the new WD
EARS which have that 4k sectors or whatever it is.
I''ve been digging through the archives more and I''m starting
to lean
towards laptop drives. I found this discussion 
http://opensolaris.org/jive/message.jspa?messageID=468269#468269 and 
contacted someone else who has used 3 WD Scorpio Black drives in a 3 way 
mirror for a year without any problems.

This is mostly going to be a backup server so I''m not too worried about
performance.
> Re the CPU, do not go low power Atom etc, go a newish Core2 duo...the power
differential at idle is bugger all and when you want to use the nas, ZFS will
make good use of the CPU. Honestly, sit down and do the calculations on power
savings of a low power cpu and you''ll see it''s better to just
not have that 5th beer on a friday - you''ll save more money that way
and be MUCH happier with your nas performance.
>
> re. cards...I use and recommend these 8-Port SUPERMICRO AOC-USASLP-L8I UIO
SAS. They are cheap on e-bay, just work and are fast. Use them.
>
> You do want alot of ram I use 8GB, but you can use 4. Ram is cheap, ZFS
loves ram, just buy 8.
>
> IMHO (and that of the best practice guide) - you should mirror the rpool
(o/s disk). Just buy 2 cheap laptop drives and when installing choose to
"mirror" them.
I can find some motherboards with 4-6 onboard sata ports. If I go with 2 
USB flash drives for a mirrored rpool do you think that would be ok? 
Performance seems to be about the same as 5400 rpm laptop drives.
> I hope that helps.

Hi,

Is it true? Any way to find it in every hierarchy?

Thanks.

Fred

> From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss-
> bounces at opensolaris.org] On Behalf Of Fred Liu
> 
> Is it true? Any way to find it in every hierarchy?
Yup.  Nope.

If you use ZFS, you make a filesystem at whatever level you need it, in
order for the .zfs directory to be available to whatever clients will be
needing it...

Thanks.
But too many file systems may be an issue for management and also normal user
cannot create file system.
I think it should go like what NetApp''s snapshot does.
It is a pity.

Thanks.

Fred
> -----Original Message-----
> From: Edward Ned Harvey [mailto:shill at nedharvey.com]
> Sent: ???, ?? 24, 2010 10:22
> To: Fred Liu; zfs-discuss at opensolaris.org
> Subject: RE: [zfs-discuss] snapshot .zfs folder can only be seen in the
> top of a file system?
> 
> > From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss-
> > bounces at opensolaris.org] On Behalf Of Fred Liu
> >
> > Is it true? Any way to find it in every hierarchy?
> 
> Yup.  Nope.
> 
> If you use ZFS, you make a filesystem at whatever level you need it, in
> order for the .zfs directory to be available to whatever clients will
> be
> needing it...
>

On Jul 23, 2010, at 7:33 PM, Fred Liu <Fred_Liu at issi.com> wrote:
> Thanks.
> But too many file systems may be an issue for management and also normal
user cannot create file system.
The ability to create or snapshot a file system can easily be delegated to a
user.
> I think it should go like what NetApp''s snapshot does.
There was a long thread on this topic earlier this year. Please see the
archives for details.
> It is a pity.
Disagree. The benefits are not as great as advertised.
  -- Richard
> 
> Thanks.
> 
> Fred
> 
>> -----Original Message-----
>> From: Edward Ned Harvey [mailto:shill at nedharvey.com]
>> Sent: ???, ?? 24, 2010 10:22
>> To: Fred Liu; zfs-discuss at opensolaris.org
>> Subject: RE: [zfs-discuss] snapshot .zfs folder can only be seen in the
>> top of a file system?
>> 
>>> From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss-
>>> bounces at opensolaris.org] On Behalf Of Fred Liu
>>> 
>>> Is it true? Any way to find it in every hierarchy?
>> 
>> Yup.  Nope.
>> 
>> If you use ZFS, you make a filesystem at whatever level you need it, in
>> order for the .zfs directory to be available to whatever clients will
>> be
>> needing it...
>> 
> 
> _______________________________________________
> zfs-discuss mailing list
> zfs-discuss at opensolaris.org
> http://mail.opensolaris.org/mailman/listinfo/zfs-discuss

> -----Original Message-----
> From: Richard Elling [mailto:richard.elling at gmail.com]
> Sent: ???, ?? 24, 2010 12:48
> To: Fred Liu
> Cc: Edward Ned Harvey; zfs-discuss at opensolaris.org
> Subject: Re: [zfs-discuss] snapshot .zfs folder can only be seen in the
> top of a file system?
> 
> 
> 
> On Jul 23, 2010, at 7:33 PM, Fred Liu <Fred_Liu at issi.com> wrote:
> 
> > Thanks.
> > But too many file systems may be an issue for management and also
> normal user cannot create file system.
> 
> The ability to create or snapshot a file system can easily be delegated
> to a user.
  
That is doable. But for normal users, they maybe don''t know ZFS and IT
architecture so much.
> 
> > I think it should go like what NetApp''s snapshot does.
> 
> There was a long thread on this topic earlier this year. Please see the
> archives for details.
Do you have the URL? I don''t have a long
subscription
> 
> > It is a pity.
> 
> Disagree. The benefits are not as great as advertised.
Is it ZFS intrinsic flaw to realize it? I have lots of users who are addicted to
NetApp''s snapshot so much!

Thanks.

Fred
>   -- Richard
> 
> >
> > Thanks.
> >
> > Fred
> >
> >> -----Original Message-----
> >> From: Edward Ned Harvey [mailto:shill at nedharvey.com]
> >> Sent: ???, ?? 24, 2010 10:22
> >> To: Fred Liu; zfs-discuss at opensolaris.org
> >> Subject: RE: [zfs-discuss] snapshot .zfs folder can only be seen
in
> the
> >> top of a file system?
> >>
> >>> From: zfs-discuss-bounces at opensolaris.org
[mailto:zfs-discuss-
> >>> bounces at opensolaris.org] On Behalf Of Fred Liu
> >>>
> >>> Is it true? Any way to find it in every hierarchy?
> >>
> >> Yup.  Nope.
> >>
> >> If you use ZFS, you make a filesystem at whatever level you need
it,
> in
> >> order for the .zfs directory to be available to whatever clients
> will
> >> be
> >> needing it...
> >>
> >
> > _______________________________________________
> > zfs-discuss mailing list
> > zfs-discuss at opensolaris.org
> > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss

>> > I think it should go like what NetApp''s snapshot does.
>>
>> There was a long thread on this topic earlier this year. Please see the
>> archives for details.
>
> Do you have the URL? I don''t have a long subscription
>>
I too do not have a long subscription, and I would be interested in
the subject line
so I can search and do further reading.

-- 

Sam Fourman Jr.
Fourman Networks
http://www.fourmannetworks.com

On 07/23/2010 02:39 AM, tomwaters wrote:
 > Re the CPU, do not go low power Atom etc, go a newish
 > Core2 duo...the power differential at idle is bugger all
 > and when you want to use the nas, ZFS will make good use
 > of the CPU.

Good advice - ZFS can use quite a lot of CPU cycles. A low-end AMD quad-core is 
another good choice here. Even better, the AMD chips support ECC RAM, though you
must get a motherboard with a BIOS that supports it. I have an Athlon II X4 630 
in my setup, and it has plenty of horsepower for an eight-disk RAIDZ2.

 > re. cards...I use and recommend these 8-Port SUPERMICRO
 > AOC-USASLP-L8I UIO SAS. They are cheap on e-bay, just work
 > and are fast. Use them.

Keep in mind you''ll have to modify the bracket to make the card fit
(unless
you''re using a Supermicro UIO case). Another good alternative with the
same
chipset is the Intel SASUC8I. This is sold as a bare card, and you need to order
the fan-out cables separately.

On Jul 24, 2010, at 01:20, Sam Fourman Jr. wrote:
>>>> I think it should go like what NetApp''s snapshot does.
>>>
>>> There was a long thread on this topic earlier this year. Please  
>>> see the
>>> archives for details.
>>
>> Do you have the URL? I don''t have a long subscription
>
> I too do not have a long subscription, and I would be interested in
> the subject line so I can search and do further reading.
The archives are available in two different interfaces:

	http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
	http://opensolaris.org/jive/forum.jspa?forumID=80

> From: Fred Liu [mailto:Fred_Liu at issi.com]
> 
> But too many file systems may be an issue for management and also
> normal user cannot create file system.
> I think it should go like what NetApp''s snapshot does.
> It is a pity.
For Windows/CIFS clients, the solution I use is:
	ln -s .zfs/snapshot snapshot
This allows people browsing a filesystem to simply see a "snapshot"
directory and browse into it.  There isn''t a user on earth
that''s too dumb
to use this.

For Windows/CIFS clients, many other people prefer the "Previous
Versions"
support, which is available if you use the "sharecifs" property (or
whatever
it is.)  Personally I prefer a complex installation of Samba instead.

For anybody using NFS/linux/unix clients, I recommend zhist.  (But I''m
biased in favor of zhist.)

I agree the ".snapshot" directory in Netapp-style is more user
friendly, and
kind of cooler, but fundamentally a security problem.  See here:
http://www.mail-archive.com/zfs-discuss at opensolaris.org/msg38057.html

> From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss-
> bounces at opensolaris.org] On Behalf Of Russ Price
> 
> Good advice - ZFS can use quite a lot of CPU cycles. A low-end AMD
> quad-core is
I know "a lot of CPU cycles" is a relative term.  But I never notice
CPU
utilization, even under the heaviest loads I can generate.  Note:  I''m
not
generally using compression (which will require CPU) and I''m not using
dedupe (which will require RAM).

Still, I don''t think it''s fair or accurate to generalize and
say "ZFS will
use a lot of CPU cycles," unless you''re qualifying it specifically
such as
"if you have compression enabled."

On 7/24/2010 8:12 PM, Edward Ned Harvey wrote:
>> From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss-
>> bounces at opensolaris.org] On Behalf Of Russ Price
>>
>> Good advice - ZFS can use quite a lot of CPU cycles. A low-end AMD
>> quad-core is
> I know "a lot of CPU cycles" is a relative term.  But I never
notice CPU
> utilization, even under the heaviest loads I can generate.  Note: 
I''m not
> generally using compression (which will require CPU) and I''m not
using
> dedupe (which will require RAM).
>
> Still, I don''t think it''s fair or accurate to generalize
and say "ZFS will
> use a lot of CPU cycles," unless you''re qualifying it
specifically such as
> "if you have compression enabled."
I was wondering about that too because I''ve been seeing a lot of zfs 
builds with atom processors and haven''t complained about cpu
utilization
given a small home server application.

Sid question... I recently ran across this blog post that indicates 
raidz and raidz2 don''t increase performance over single drive 
performance unlike raid5. The post was old so I was wondering if that 
was still true. 
http://blogs.sun.com/relling/entry/zfs_raid_recommendations_space_performance

On Jul 24, 2010, at 12:53 AM, Fred Liu wrote:
>> 
>> Disagree. The benefits are not as great as advertised.
> 
> Is it ZFS intrinsic flaw to realize it? I have lots of users who are
addicted to NetApp''s snapshot so much!
The tricky part is exporting FIDs over NFS in a way that allows you to
''cd ..'' from any .zfs direcotry (or snapshot within
.zfs/snapshots) and end up back where you started, because the snapshots are
exported as part of the parent filesystem, even though under the hood they are
different vfs_t structures.

I''d encourage you to work on the RFE yourself - any implementation
would certainly be appreciated.  This possibility was originally why the
''snapdir'' property was named as it was, so we could someday
support ''snapdir=every'' to export .zfs in every directory.

- Eric

--
Eric Schrock, Fishworks                        http://blogs.sun.com/eschrock

On Jul 24, 2010, at 5:37 PM, JavaWebDev wrote:
> On 7/24/2010 8:12 PM, Edward Ned Harvey wrote:
>>> From: zfs-discuss-bounces at opensolaris.org [mailto:zfs-discuss-
>>> bounces at opensolaris.org] On Behalf Of Russ Price
>>> 
>>> Good advice - ZFS can use quite a lot of CPU cycles. A low-end AMD
>>> quad-core is
>> I know "a lot of CPU cycles" is a relative term.  But I never
notice CPU
>> utilization, even under the heaviest loads I can generate.  Note: 
I''m not
>> generally using compression (which will require CPU) and I''m
not using
>> dedupe (which will require RAM).
>> 
>> Still, I don''t think it''s fair or accurate to
generalize and say "ZFS will
>> use a lot of CPU cycles," unless you''re qualifying it
specifically such as
>> "if you have compression enabled."
> 
> I was wondering about that too because I''ve been seeing a lot of
zfs builds with atom processors and haven''t complained about cpu
utilization given a small home server application.
> 
> Sid question... I recently ran across this blog post that indicates raidz
and raidz2 don''t increase performance over single drive performance
unlike raid5. The post was old so I was wondering if that was still true.
http://blogs.sun.com/relling/entry/zfs_raid_recommendations_space_performance
This model is for small, random read performance only.

I''m perfectly happy to entertain models which can accurately predict
performance when caches are present.  But, I''ve never seen one that
is accurate :-(
 -- richard 

-- 
ZFS and performance consulting
http://www.RichardElling.com

I''ve been looking at using consumer 2.5" drives also, I think the
ones I''ve settled on are the hitachi 7K500 500 GB. These are 7200 rpm,
I''m concerned the 5400''s might be a little too low performance
wise. The main reasons for hitachi were performance seems to be among the top 2
or 3 in the laptop drive segment, I''ve found hitachi to be pretty
reliable, and perhaps most importantly is there is the hitachi feature tool,
which allows you to disable the head unload feature. You don''t need to
set it on each reboot, plus it''s persistent across reboots.
-- 
This message posted from opensolaris.org