Current best practices for new pools, particularly wrt right

Here you can discuss every aspect of OpenZFS on OS X. Note: not for support requests!

Current best practices for new pools, particularly wrt right

Postby leeb » Thu Nov 19, 2015 10:06 pm

At long, long last I'm making the move away from ZEVO/10.8 (that sure took a lot longer then I expected!) straight to 10.11 on my main systems. I'm excited about all that has happened with OpenZFS, but was wondering what the current best practices are, particularly with regard to right-sizing (also known as short stroking). That was originally an issue brought up way back when ZFS was still with Sun, wherein drives from different manufacturers that were nominally the same size had different sector counts despite there being some industry standards for that. Ie., for one of my main pools I'll be using "6TB" drives, which actually equates to 11,721,045,168 (IDEMA standard) 512-byte blocks, or exactly 6,001,175,126,016 bytes. The issue in the past was that as BPR never happened if I ever needed to replace as disk but it was even a single block smaller despite the same nominal size it'd be a no-go. In turn the easiest way to deal with that was to simply chop it down a bit, giving up an insignificant chunk out of 6000 but adding a bit of extra resiliency for any future replacement. I did in fact see this in practice a few times under Solaris, but that was ages ago and I don't know if it ever happens anymore. Do all manufacturers adhere to the IDEMA sector formula now and have identical sector counts? If right sizing is still something considered, is it possible to still get the benefits of giving ZFS a whole disk, perhaps via setting the whole_disk label manually (or perhaps booting linux and using hdparm for an HPA)? The intention is still for ZFS to have the entire disk under management with all optimizations that go with that, just not utilizing 100% of the physical block count. If that's not even worth bothering with at this point though all the better.

Anything else worth considering beyond the standard stuff like lz4? As these pools won't be moving between OpenZFS implementations directly I'm assuming I don't need to worry about feature flag compatibility right now. I have seen conflicting advice regarding ashift performance of pools using only SSD vdevs. Some posts suggest just using the smallest LBA the SSD offers (512 or 4k), while the OpenZFS main wiki Hardware page suggests matching page size. In a non-server situation though with SSDs I'm not sure the differences will even matter much. Finally, looks like APM is still a problem, I guess mechanical drives just plain stopped entering into Apple's consciousness at all a long time ago.

Thanks for any thoughts (and all the hard work from everyone who have made O3X happen at all)! It'll be interesting to get everything running again and see how the new system handles odd home layouts. 10.8 had some major quirks in terms of what the Finder would tolerate (for example ~ could be its own filesystem, and so could all subfolders, except ~/Library could not be at all and Desktop would only work symlinked).
leeb
 
Posts: 8
Joined: Thu May 15, 2014 12:10 pm

Re: Current best practices for new pools, particularly wrt r

Postby lundman » Mon Nov 23, 2015 4:47 pm

My memory is a bit fuzzy, but I recall a set of patches done, which rounded off the last few megs on disks, so that we can avoid that whole "device is too small" problem. At least I've not come across it since 2009ish Solaris days.
User avatar
lundman
 
Posts: 311
Joined: Thu Mar 06, 2014 2:05 pm
Location: Tokyo, Japan

Re: Current best practices for new pools, particularly wrt r

Postby leeb » Wed Nov 25, 2015 1:21 pm

lundman wrote:My memory is a bit fuzzy, but I recall a set of patches done, which rounded off the last few megs on disks, so that we can avoid that whole "device is too small" problem. At least I've not come across it since 2009ish Solaris days.

Thank you, hunting around I did find some references to that though I couldn't trace down a direct commit. After significant consolidation the last half decade there are also just plain few manufacturers of good old spinning rust remaining, so maybe everyone just sticks to IDEMA anyway.

At any rate good enough and thank you for the reply.
leeb
 
Posts: 8
Joined: Thu May 15, 2014 12:10 pm


Return to General Discussions

Who is online

Users browsing this forum: No registered users and 1 guest

cron