--- On Fri, 2/8/13, Gordon Ross <[email protected]> wrote:
> From: Gordon Ross <[email protected]> > Subject: Re: [discuss] The cylinder model & 4k sectors > To: [email protected] > Date: Friday, February 8, 2013, 11:12 AM > On Fri, Feb 8, 2013 at 11:42 AM, > Reginald Beardsley <[email protected]> > wrote: > > > > > > --- On Fri, 2/8/13, Gary Driggs <[email protected]> > wrote: > > > >> From: Gary Driggs <[email protected]> > >> Subject: Re: [discuss] The cylinder model & 4k > sectors > >> To: "[email protected]" > <[email protected]> > >> Date: Friday, February 8, 2013, 9:53 AM > >> On Feb 8, 2013, at 7:35 AM, Reginald > >> Beardsley wrote: > >> > >> > That is probably going to get torn down and a > 4th disk > >> added so I can run /export on RAIDZ2 and be able to > take a 2 > >> disk failure on a 4 disk system w/o loss of data > >> > >> Have you considered the option of installing the OS > on a > >> server grade USB drive? > >> > > > > I assume you mean "thumb drive". I rejected that > because it's a single point of failure. There's no guarantee > that it wouldn't corrupt the RAIDZ pool during a failure. > > > > A 10 GB rpool in s0 is 0.5% of a 2 TB disk. So > it's under $3 to have a 4-way mirrored rpool to support a > 3.5+ TB RAIDZ2 /export. That's cheaper than a USB > flash and more robust. > > > > Have Fun! > > Reg > > > I suggest you have a look at this prior work: > > 2949 format(1m) should support megabyte-aligned partitions > http://www.yalms.org/cr/illumos-2949/ > > There were some discussions of it about 6 months ago on > developers > (search for issue 2949) > http://www.listbox.com/member/archive/182179/sort/time_rev/?search_for=2949 > > Basically, that was a start on the problem, but I missed > some of the places that know about geometry. > The fix needs to "hit them all". > > Sorry, I don't have much time to work on this right now, > so I'd welcome you or anyone else taking it over. I've looked at all the references, but I'm a bit unclear where the problem was. Where is the "cmlb" code? I gather there is a bunch of study needed to determine what happens there if the change you made to format(1m) is made. I know that fmthard(1m) is also affected for SMI label disks and assume there is probably more. I am *not* familiar w/ Solaris internals. I've got the book (1st & 2nd ed) by McDougall & Mauro, but have not read an internals book since the 4.4 BSD book came out. I started reading the FreeBSD book when it came out but ran out of steam as other interests were taking hold. The last OS book I read was probably Helios, though I did keep buying them as they came out. So I'm likely to have any references I might need. I've also got a mercurial copy of illumos-gate. 20 years ago I was intensely interested in OS internals, so I knew a lot about it at one time. However I doubt I remember much of anything after such a long time and I never did take the time to learn SVr4 and Solaris beyond the userland interface and a very rudimentary notion of the internal structure. So no promises, but if people will point me in the right directions I'll see what I can do. I can read quickly, but finding what to read can be more of a challenge. The index to "Solaris Internals" does not mention cmlb. >From the discussion, this is kernel code which I was hoping to stay out of >because the paging required. But this beat me up badly enough I'd like to get >revenge and fix it. I'm subscribed to this and the dev list, but don't know what the convention is for what gets discussed where, so I'll follow whatever lead anyone wishes to offer. Have Fun! Reg ------------------------------------------- illumos-discuss Archives: https://www.listbox.com/member/archive/182180/=now RSS Feed: https://www.listbox.com/member/archive/rss/182180/21175430-2e6923be Modify Your Subscription: https://www.listbox.com/member/?member_id=21175430&id_secret=21175430-6a77cda4 Powered by Listbox: http://www.listbox.com
