Hi, I guess we are acquainted with the ZFS Wikipedia? http://en.wikipedia.org/wiki/ZFS Customers refer to it, I wonder where the Wiki gets its numbers. For example there''s a Sun marketing slide that says "unlimited snapshots" contradicted by the the first bullet: 2^48 ? Number of snapshots in any file system (2 ? 10^14) Is this correct? I want to make a mirrored pool but after creation detach one side of the mirror and run like this indefinitely. Can I do this? I have yet to try it out, just wondering if someone knows the answer right off the bat. Why can''t I have a vdev of type concat? A customer says he has hardware RAID-0 and does not want to stripe over stripe. Are there known problems with stripes on stripes, unintentional hotspots maybe? Data checksum is used only to detect bad data, and not to recover data, correct? Just verifying. Can I use ZFS specifically with PowerPath? I remember George W. mentioning that PowerPath tries to do something clever and it interfered with ZFS but I can''t remember the specifics. Will I be able to I tune the DMU "flush" rate, now set at 5 seconds? CT
Hello Christine,
Saturday, December 16, 2006, 12:17:12 AM, you wrote:
CT> Hi,
CT> I guess we are acquainted with the ZFS Wikipedia?
CT> http://en.wikipedia.org/wiki/ZFS
CT> Customers refer to it, I wonder where the Wiki gets its numbers. For
CT> example there''s a Sun marketing slide that says "unlimited
snapshots"
CT> contradicted by the the first bullet:
CT> 2^48 ? Number of snapshots in any file system (2 ? 10^14)
CT> Is this correct?
I belive it is.
However 2^48 limit per file system is for every practical purpose
virtually no limit, right? So when it comes to marketing I guess they
get it right.
CT> I want to make a mirrored pool but after creation detach one side of the
CT> mirror and run like this indefinitely. Can I do this? I have yet to
CT> try it out, just wondering if someone knows the answer right off the bat.
Of course you can.
However you won''t be able to "mount" detached submirror.
I guess it should be quite easy to implement if needed (yes, it''s
needed).
CT> Why can''t I have a vdev of type concat? A customer says he has
hardware
CT> RAID-0 and does not want to stripe over stripe. Are there known
CT> problems with stripes on stripes, unintentional hotspots maybe?
Depends how those stripes on HW array are actually configured and
depends on data access pattern. Once you put enough data that more
LUNs are used then even with concatenation you''ll get into trouble
(assuming pathological situation already). But generally if config is
done right stripe on stripe shouldn''t be a problem at all (rather
should give you more performance).
By pathological situation I assume config like creating a stripe from
some hypers (partitions) each on differen disk, then making another
stripe from other hypers on the same disks and then striping between
such LUNs on a host.
CT> Data checksum is used only to detect bad data, and not to recover data,
CT> correct? Just verifying.
Depends. If you do only striping in ZFS then you''re right (not
counting ditto blocks for meta data - however it can be useful).
Once you create redundant pool in ZFS then you''re not only able to
detect bad data but also to correct them on the fly. That''s what ZFS
does automatically for you.
CT> Can I use ZFS specifically with PowerPath? I remember George W.
CT> mentioning that PowerPath tries to do something clever and it interfered
CT> with ZFS but I can''t remember the specifics.
In theory it should work. However someone else reported a problem with
such a combination and IIRC it was about not exporting all LUN details
by PP rather than doing anything clever by PP itself. I hope it was
fixed by EMC and if not perhaps it''s a good idea to fill in BUG in EMC
(maybe I should test it and do it my self - just need some time).
CT> Will I be able to I tune the DMU "flush" rate, now set at 5
seconds?
echo ''txg_time/D 0t1'' | mdb -kw
Will change it to every 1s.
--
Best regards,
Robert mailto:rmilkowski at task.gda.pl
http://milek.blogspot.com
> CT> Will I be able to I tune the DMU "flush" rate, now set at 5 seconds? > > echo ''txg_time/D 0t1'' | mdb -kwEr, that ''D'' ahould be a ''W''. Having said that I don''t think we recommend messing with the transaction group commit timing.
Hello Neil, Monday, December 18, 2006, 5:48:40 AM, you wrote:>> CT> Will I be able to I tune the DMU "flush" rate, now set at 5 seconds? >> >> echo ''txg_time/D 0t1'' | mdb -kwNP> Er, that ''D'' ahould be a ''W''. NP> Having said that I don''t think we recommend messing with the transaction NP> group commit timing. Yep, W - sorry for that. -- Best regards, Robert mailto:rmilkowski at task.gda.pl http://milek.blogspot.com
Neil Perrin wrote:> Having said that I don''t think we recommend messing with the transaction > group commit timing.Yeah I don''t think the customer means to tune it this way either, they were thinking of something like tune_t_fsflushr (is this still in use?) They want to know when the txg timing is a tunable. No takers for the other questions? The data checksum question is with regards to how to recover, customer wants to make sure it''s not like you can do something with the checksum to get back the data (like a 1-bit detection + recovery but 2-bit detection but no recovery). I don''t think it works this way but just have to check. And the PowerPath question is important, customer is using PP right now. CT
Additional comments below... Christine Tran wrote:> Hi, > > I guess we are acquainted with the ZFS Wikipedia? > http://en.wikipedia.org/wiki/ZFS > > Customers refer to it, I wonder where the Wiki gets its numbers. For > example there''s a Sun marketing slide that says "unlimited snapshots" > contradicted by the the first bullet: > > 2^48 ? Number of snapshots in any file system (2 ? 10^14) > > Is this correct? > > I want to make a mirrored pool but after creation detach one side of the > mirror and run like this indefinitely. Can I do this? I have yet to > try it out, just wondering if someone knows the answer right off the bat. > > Why can''t I have a vdev of type concat? A customer says he has hardware > RAID-0 and does not want to stripe over stripe. Are there known > problems with stripes on stripes, unintentional hotspots maybe?Friends don''t let friends use RAID-0. Everyone I''ve known who insists on doing so has a sad story to tell. ZFS does not do concat, it does dynamic striping, which has the benefits of RAID-0 with the ability to easily expand like concat.> Data checksum is used only to detect bad data, and not to recover data, > correct? Just verifying.Yes, checksum is part of the detection mechanism. Correction is done by several other mechanisms, primarily using redundancy or parity reconstruction. -- richard> Can I use ZFS specifically with PowerPath? I remember George W. > mentioning that PowerPath tries to do something clever and it interfered > with ZFS but I can''t remember the specifics. > > Will I be able to I tune the DMU "flush" rate, now set at 5 seconds? > > CT > _______________________________________________ > zfs-discuss mailing list > zfs-discuss at opensolaris.org > http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
Christine Tran wrote:> > > And the PowerPath question is important, customer is using PP right now. >I haven''t heard any powerpath issues. Can you track down what it was GeorgeW mentioned?
Torrey McMahon wrote:> I haven''t heard any powerpath issues. Can you track down what it was > GeorgeW mentioned?Well, the problem is I can''t remember. It was during a ZFS TOI class, and perhaps it was that PP tries to be clever by grouping tsx together ... If there''s been no PP issue reported, that''s what I''m going to have to go on, unless I have cases to the contrary. CT
Hello Torrey, Monday, December 18, 2006, 8:38:42 PM, you wrote: TM> Christine Tran wrote:>> >> >> And the PowerPath question is important, customer is using PP right now. >>TM> I haven''t heard any powerpath issues. Can you track down what it was TM> GeorgeW mentioned? Hmmmm... http://www.webservertalk.com/archive100-2006-9-1594527.html It looks like IBM''s software is problematic and PP works. I''m almost sure there was some problem with PP but looks like it''s working (now). -- Best regards, Robert mailto:rmilkowski at task.gda.pl http://milek.blogspot.com