[OpenIndiana-discuss] sd.conf trouble and illumos bug #3220

[Bryan N Iotti]

Adding a wildcard for the all devices to use the 4K block size did indeed work and drastically improved the performance (and, I guess, life expectancy) of that SSD.

I also added "physical-block-size=512" entries for all traditional 512-byte block disks connected to this machine and they are running fine.

The scrub speed on the SSD (only quick test I could think of to try sustained read and write) went from 160MB/sec to 250MB/sec. If I use 8K it doesn't change as much, but seems to be more constant. System has been running for a couple of hours and it's nice and snappy.

Should I add a line to the OI wiki about the wildcard options? I didn't see them there last time I looked and they can be handy for newcomers and the like.

Bryan

On Mon, 6 Jan 2014 12:49:53 -0800 (PST)
Reginald Beardsley <pulaskite at yahoo.com> wrote:

> Sadly yahoo mail makes a mess of replies embedded in the text :-(
> 
> I'll have to try the 3 TB USB drive again.  I certainly can't see any reason that your wildcarding scheme shouldn't work.  If there is a problem it's probably a bug.
> 
> There are so many lies being told about geometry that it's hard to say what you should do.  At one point I wound up with a partition that was not 4k aligned if I started with cylinder #1, so I used #2 which was 4k aligned.  However, the 2k aligned partition seemed to work OK, so I'm not certain.  I wasn't willing to leave it that way to avoid wasting a little space.
> 
> I can't think of any reason that 512B drives wouldn't work fine aligned to larger boundaries.   The only consequence I can see is some wasted space which really isn't an issue given we're talking pennies per GB.  I can't imagine a performance issue that could result.
> 
> It seems to me that the default alignment rules would give less trouble if the alignment was to the most restrictive requirement.
> 
> --------------------------------------------
> On Mon, 1/6/14, Bryan N Iotti <ironsides.medvet at runbox.com> wrote:
> 
>  Subject: Re: [OpenIndiana-discuss] sd.conf trouble and illumos bug #3220
>  To: openindiana-discuss at openindiana.org
>  Date: Monday, January 6, 2014, 12:08 PM
>  
>  Hi Reginald,
>  
>  Thanks for your reply.
>  
>  In the end, I just booted from a live DVD and plugged that
>  drive in on my laptop using a double asterisk as the sd.conf
>  line and an 8k block size. A test pool create read back
>  ashift=13, which should be fine.
>  
>  Then I used format -e and fdisk to create a single Solaris
>  slice with start and end blocks that would fall on a
>  cylinder number that was divisible by 8 (I did it like for
>  4K disks, if I'm wrong I'd like some input on that please).
>  
>  Then I created the rpool in the first slice and moved the
>  disk over to the final system, where I sent and received the
>  zfs datasets. Change menu.lst, install grub and I was done.
>  Working fine so far (160MB/sec sustained on a scrub, normal
>  wsvc_time and asvc_time in iostat).
>  
>  Would anyone see any problem in doing the sd.conf line
>  backwards, I mean masking all drives as "**" with a 4K
>  blocksize and adding entries for the other 512-byte drives?
>  What would happen if ZFS thought a 512-byte block device
>  used a 4K block size? Wasted space, performance issue,
>  errors?
>  
>  Thanks, as always, for any and all input.
>  
>  
>     Bryan
>  
>  
>  On Sun, 5 Jan 2014 16:10:58 -0800 (PST)
>  Reginald Beardsley <pulaskite at yahoo.com>
>  wrote:
>  
>  > I can't comment on your particular issue which is why I
>  didn't respond earlier.  However,  I had trouble
>  getting a 3 TB Toshiba USB drive to work properly.  I
>  spent a bunch of time reading the code that parses sd.conf,
>  but finally gave up and didn't go further.
>  > 
>  > On reflection I wonder if the problem is a failure to
>  propagate the information about block size to ZFS correctly
>  under certain circumstances.  I think it would be nice
>  if one could force ashift when creating a pool. 
>  Automagic is nice, but there's often no substitute for human
>  intelligence.  I spent a couple of days trying to
>  persuade OI to create a properly aligned pool before I gave
>  up.
>  > 
>  > 
>  >  
>  > 
>  > _______________________________________________
>  > OpenIndiana-discuss mailing list
>  > OpenIndiana-discuss at openindiana.org
>  > http://openindiana.org/mailman/listinfo/openindiana-discuss
>  > 
>  
>  
>  -- 
>  Bryan N Iotti
>  
>  +39 366 3708436
>  ironsides.medvet at runbox.com
>  
>  _______________________________________________
>  OpenIndiana-discuss mailing list
>  OpenIndiana-discuss at openindiana.org
>  http://openindiana.org/mailman/listinfo/openindiana-discuss
>  
> 
> _______________________________________________
> OpenIndiana-discuss mailing list
> OpenIndiana-discuss at openindiana.org
> http://openindiana.org/mailman/listinfo/openindiana-discuss
> 


-- 
Bryan N Iotti

+39 366 3708436
ironsides.medvet at runbox.com