Re: OpenBSD's FFS/dirpref/softupdates improvements
On Thu, Apr 19, 2001 at 08:57:37AM -0700, Jeremiah Gowdy wrote: "Two aspects of the FFS filesystem in OpenBSD have received significant improvements since 2.8, increasing performance dramatically. Thanks to art, gluk, csapuntz, and a host of other developers and testers, Soft Updates are now much more stable than ever before. The second improvement, contributed by [EMAIL PROTECTED], is a new directory allocation policy (codenamed "dirpref"). Coupled with soft updates, the new dirpref code offers up to a 60x speed increase in gluk's tests, documented here:" http://groups.google.com/groups?q=dirprefnum=100hl=enlr=safe=offrnum=2; seld=905073910ic=1 Does anyone know anything about this ? Commited to -current about 10 April. I suspect that Jordan would shoot someone who suggested a MFC before 4.3 is out. -- Mike Bristow, seebitwopie To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: OpenBSD's FFS/dirpref/softupdates improvements
On 19-Apr-01 Mike Bristow wrote: On Thu, Apr 19, 2001 at 08:57:37AM -0700, Jeremiah Gowdy wrote: "Two aspects of the FFS filesystem in OpenBSD have received significant improvements since 2.8, increasing performance dramatically. Thanks to art, gluk, csapuntz, and a host of other developers and testers, Soft Updates are now much more stable than ever before. The second improvement, contributed by [EMAIL PROTECTED], is a new directory allocation policy (codenamed "dirpref"). Coupled with soft updates, the new dirpref code offers up to a 60x speed increase in gluk's tests, documented here:" http://groups.google.com/groups?q=dirprefnum=100hl=enlr=safe=offrnum=2; seld=905073910ic=1 Does anyone know anything about this ? Commited to -current about 10 April. I suspect that Jordan would shoot someone who suggested a MFC before 4.3 is out. It needs more work, too. If you try to use an old fsck with the new kernel, then the old fsck will clobber some new variables in the superblock. Then the new kernel will panic later on instead of doing a sanity check on the new values in the superblock and falling back to defaults if they are bogus. This is a major POLA bug and the changes shouldn't go into 4.x unless this is fixed. This breaks the recommended method of updating stable by doing the installworld after rebooting into a new kernel. -- John Baldwin [EMAIL PROTECTED] -- http://www.FreeBSD.org/~jhb/ PGP Key: http://www.baldwin.cx/~john/pgpkey.asc "Power Users Use the Power to Serve!" - http://www.FreeBSD.org/ To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: OpenBSD's FFS/dirpref/softupdates improvements
I've accepted the job of MFCing the dirpref stuff to -stable ... after the 4.3 release. If fsck is clobbering consistently we can probably make the kernel avoid a panic. I'll look at the issue carefully when I do the MFC. -Matt :On 19-Apr-01 Mike Bristow wrote: : On Thu, Apr 19, 2001 at 08:57:37AM -0700, Jeremiah Gowdy wrote: : "Two aspects of the FFS filesystem in OpenBSD have received significant : improvements since 2.8, increasing performance dramatically. Thanks to art, : gluk, csapuntz, and a host of other developers and testers, Soft Updates are : now much more stable than ever before. The second improvement, contributed : by [EMAIL PROTECTED], is a new directory allocation policy (codenamed : "dirpref"). Coupled with soft updates, the new dirpref code offers up to a : 60x speed increase in gluk's tests, documented here:" : : http://groups.google.com/groups?q=dirprefnum=100hl=enlr=safe=offrnum=2; : seld=905073910ic=1 : : Does anyone know anything about this ? : : Commited to -current about 10 April. : : I suspect that Jordan would shoot someone who suggested a MFC before : 4.3 is out. : :It needs more work, too. If you try to use an old fsck with the new kernel, :then the old fsck will clobber some new variables in the superblock. Then the :new kernel will panic later on instead of doing a sanity check on the new :values in the superblock and falling back to defaults if they are bogus. This :is a major POLA bug and the changes shouldn't go into 4.x unless this is fixed. :This breaks the recommended method of updating stable by doing the installworld :after rebooting into a new kernel. : :-- : :John Baldwin [EMAIL PROTECTED] -- http://www.FreeBSD.org/~jhb/ To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: OpenBSD's FFS/dirpref/softupdates improvements
On 19-Apr-01 Matt Dillon wrote: I've accepted the job of MFCing the dirpref stuff to -stable ... after the 4.3 release. If fsck is clobbering consistently we can probably make the kernel avoid a panic. I'll look at the issue carefully when I do the MFC. -Matt The problem lies in that fsck detects that the dirpref tunables differ between the alternate superblock and the first superblock, so it prompts you to overwrite the first superblock with the values in the alternate. For some reason the values in the alternate end up panic'ing the system. fsck -p doesn't exhibit this problem, IIRC, only an actual fsck or fsck -y. Having the kernel sanity check these parameters in the superblock should be sufficient. :On 19-Apr-01 Mike Bristow wrote: : On Thu, Apr 19, 2001 at 08:57:37AM -0700, Jeremiah Gowdy wrote: : "Two aspects of the FFS filesystem in OpenBSD have received significant : improvements since 2.8, increasing performance dramatically. Thanks to : art, : gluk, csapuntz, and a host of other developers and testers, Soft Updates : are : now much more stable than ever before. The second improvement, contributed : by [EMAIL PROTECTED], is a new directory allocation policy (codenamed : "dirpref"). Coupled with soft updates, the new dirpref code offers up to a : 60x speed increase in gluk's tests, documented here:" : : http://groups.google.com/groups?q=dirprefnum=100hl=enlr=safe=offrnum=2 : : seld=905073910ic=1 : : Does anyone know anything about this ? : : Commited to -current about 10 April. : : I suspect that Jordan would shoot someone who suggested a MFC before : 4.3 is out. : :It needs more work, too. If you try to use an old fsck with the new kernel, :then the old fsck will clobber some new variables in the superblock. Then :the :new kernel will panic later on instead of doing a sanity check on the new :values in the superblock and falling back to defaults if they are bogus. :This :is a major POLA bug and the changes shouldn't go into 4.x unless this is :fixed. :This breaks the recommended method of updating stable by doing the :installworld :after rebooting into a new kernel. : :-- : :John Baldwin [EMAIL PROTECTED] -- http://www.FreeBSD.org/~jhb/ -- John Baldwin [EMAIL PROTECTED] -- http://www.FreeBSD.org/~jhb/ PGP Key: http://www.baldwin.cx/~john/pgpkey.asc "Power Users Use the Power to Serve!" - http://www.FreeBSD.org/ To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: OpenBSD's FFS/dirpref/softupdates improvements
On Fri, 20 Apr 2001, Boris Popov wrote: On Thu, 19 Apr 2001, John Baldwin wrote: It needs more work, too. If you try to use an old fsck with the new kernel, then the old fsck will clobber some new variables in the superblock. Then the new kernel will panic later on instead of doing a sanity check on the new values in the superblock and falling back to defaults if they are bogus. This is a major POLA bug and the changes shouldn't go into 4.x unless this is fixed. This breaks the recommended method of updating stable by doing the installworld after rebooting into a new kernel. On other hand, it makes some troubles if filesystem is shared between 4.x and -current :) -- Boris Popov Hm, something just struck me. a) Why is fsck checking fields that it doesn't know as anything other than filler? b) When it says that the alternate superblock doesn't match, is that really true? If so, why isn't that being updated? Of course the kernel should sanity check the values so it doesn't panic, but making fsck not louse up future filesystem changes seems prudent as well. (If that's feasible.) Mike "Silby" Silbersack To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
