Обсуждение: Win32 port powerfail testing
Despite some people's thoughts that a powerfail test is of little use, I going to spend some time doing one anyway because I think Tom's arguments for it are valid. I have lashed together the attached test program (the important bits are the setup, run and check functions) for review before I actually do anything next week. Comments, suggestions etc are welcome, though I don't have the time to write anything too complex, but do want to perform a valid test first time round if possible. I intend to run the tests on a Dual PIII 1GHz box, with 1Gb of Non-ECC RAM and a 20Gb (iirc) IDE disk. I will run on Windows 2000 Server with an NTFS filesystem, and again on Slackware Linux 8 with either ext3 or reiserfs (which is preferred?). The number of runs will be dictated by my workload next week, but I'd like to do at least 20 powerfails on each OS. Regards, Dave.
Вложения
Dave Page kirjutas R, 31.01.2003 kell 22:36: > Despite some people's thoughts that a powerfail test is of little use, I > going to spend some time doing one anyway because I think Tom's > arguments for it are valid. I have lashed together the attached test > program (the important bits are the setup, run and check functions) for > review before I actually do anything next week. Comments, suggestions > etc are welcome, though I don't have the time to write anything too > complex, but do want to perform a valid test first time round if > possible. > > I intend to run the tests on a Dual PIII 1GHz box, with 1Gb of Non-ECC > RAM and a 20Gb (iirc) IDE disk. I will run on Windows 2000 Server with > an NTFS filesystem, and again on Slackware Linux 8 with either ext3 or > reiserfs (which is preferred?). I think that ext3 should be more reliable, or at least more mainstream - I have had bad experience with raiserfs not too long ago - a crash (similar to pull-the-plug) zeroed out completely unrelated files (files not recently written to, just read). As I don't use Slackware anymore (though I started usin linux on it in the dark ages before 1.0 kernel), I don't know if the issues are fixed there. > The number of runs will be dictated by my workload next week, but I'd > like to do at least 20 powerfails on each OS. Don't post if you happen to get better results for win32 ;) I have a worried lung-doctor (aka pulmonologist) friend who did some research/statistics on influence of smoking and he is desperate as his methodologically completely scientific studies ended up showing that smoking is healthy and not smoking is not ;), so he seems unable to publish any of his results in any respectable outlet ;( -- Hannu Krosing <hannu@tm.ee>
On Fri, 2003-01-31 at 14:36, Dave Page wrote: > > I intend to run the tests on a Dual PIII 1GHz box, with 1Gb of Non-ECC > RAM and a 20Gb (iirc) IDE disk. I will run on Windows 2000 Server with > an NTFS filesystem, and again on Slackware Linux 8 with either ext3 or > reiserfs (which is preferred?). > Please go with XFS or ext3. There are a number of blessed and horror stories which still float around about reiserfs (recent and old; even though I've never lost data with it -- using it now even). Might be worth testing FAT32 on NT as well. Even if we don't advocate it's use, it may not hurt to at least get an understanding of what one might reasonably expect from it. I'm betting there are people just waiting to run with FAT32 in the Win32 world. ;) Regards, -- Greg Copeland <greg@copelandconsulting.net> Copeland Computer Consulting
On Sat, Feb 01, 2003 at 12:27:31AM -0600, Greg Copeland wrote: > On Fri, 2003-01-31 at 14:36, Dave Page wrote: > > > > I intend to run the tests on a Dual PIII 1GHz box, with 1Gb of Non-ECC > > RAM and a 20Gb (iirc) IDE disk. I will run on Windows 2000 Server with > > an NTFS filesystem, and again on Slackware Linux 8 with either ext3 or > > reiserfs (which is preferred?). > > > > Please go with XFS or ext3. There are a number of blessed and horror > stories which still float around about reiserfs (recent and old; even > though I've never lost data with it -- using it now even). > > Might be worth testing FAT32 on NT as well. Even if we don't advocate > it's use, it may not hurt to at least get an understanding of what one > might reasonably expect from it. I'm betting there are people just > waiting to run with FAT32 in the Win32 world. ;) You'd better go with NTFS. There are a number of blessed and horror stories which still float around about FAT32 (recent and old; even though I've never lost data with it -- using it now even now. Might be worth testing reiserfs on Linux as well. Even if we don't advocate it's use, it may not hurt to at least get an understanding of what one my reasonbly expect from it. I'm better there are people just waiting to run with reiserfs in the Linux world. ;) Regards, and tongue firmly in cheek, -- Adam Haberlach | "If I have hacked deeper than them, it is adam@mediariffic.com | because I stand in their trenches." http://mediariffic.com | -- Graham Nelson
Try it with FreeBSD's UFS and FreeBSD 5.0's new UFS2 filesystems perhaps - or I could! Chris On 1 Feb 2003, Greg Copeland wrote: > On Fri, 2003-01-31 at 14:36, Dave Page wrote: > > > > I intend to run the tests on a Dual PIII 1GHz box, with 1Gb of Non-ECC > > RAM and a 20Gb (iirc) IDE disk. I will run on Windows 2000 Server with > > an NTFS filesystem, and again on Slackware Linux 8 with either ext3 or > > reiserfs (which is preferred?). > > > > Please go with XFS or ext3. There are a number of blessed and horror > stories which still float around about reiserfs (recent and old; even > though I've never lost data with it -- using it now even). > > Might be worth testing FAT32 on NT as well. Even if we don't advocate > it's use, it may not hurt to at least get an understanding of what one > might reasonably expect from it. I'm betting there are people just > waiting to run with FAT32 in the Win32 world. ;) > > > Regards, > > -- > Greg Copeland <greg@copelandconsulting.net> > Copeland Computer Consulting > > > ---------------------------(end of broadcast)--------------------------- > TIP 2: you can get off all lists at once with the unregister command > (send "unregister YourEmailAddressHere" to majordomo@postgresql.org) >
On Sat, 2003-02-01 at 00:34, Adam Haberlach wrote: > On Sat, Feb 01, 2003 at 12:27:31AM -0600, Greg Copeland wrote: > > On Fri, 2003-01-31 at 14:36, Dave Page wrote: > > > > > > I intend to run the tests on a Dual PIII 1GHz box, with 1Gb of Non-ECC > > > RAM and a 20Gb (iirc) IDE disk. I will run on Windows 2000 Server with > > > an NTFS filesystem, and again on Slackware Linux 8 with either ext3 or > > > reiserfs (which is preferred?). > > > > > > > Please go with XFS or ext3. There are a number of blessed and horror > > stories which still float around about reiserfs (recent and old; even > > though I've never lost data with it -- using it now even). > > > > Might be worth testing FAT32 on NT as well. Even if we don't advocate > > it's use, it may not hurt to at least get an understanding of what one > > might reasonably expect from it. I'm betting there are people just > > waiting to run with FAT32 in the Win32 world. ;) > > You'd better go with NTFS. There are a number of blessed and horror > stories which still float around about FAT32 (recent and old; even though > I've never lost data with it -- using it now even now. > > Might be worth testing reiserfs on Linux as well. Even if we don't > advocate it's use, it may not hurt to at least get an understanding of > what one my reasonably expect from it. I'm better there are people > just waiting to run with reiserfs in the Linux world. ;) > > Regards, and tongue firmly in cheek, Touche! :P While I understand and even appreciate the humor value, I do believe the picture is slightly different than your "analysis". If we make something that runs on Win32 platforms, might it also run on Win98, WinME, etc.? Let's face the facts that should it also run on these platforms, it's probably only a matter of time before someone has it running on FAT32 (even possible on NT, etc). In other words, I'm fully expecting the lowest common denominator of MySQL user to be looking at PostgreSQL on Win32. Which potentially means lots of FAT32 use. And yes, even for a "production" environment. Ack! Double-ack! Also, Dave was asking for feedback between reiserfs and ext3. I offered XFS and ext3 as candidates. I personally believe that ext3 and XFS are going to be the more common (in that order) of journaled FS for DB Linux users. Besides, aside from any bugs in reiserfs, testing results for ext3 or XFS should probably coincide with reasonable expectations for reiserfs as well. As I consider FAT32 to be much more fragile than ext2 (having had seriously horrendous corruption and repaired/recovered from it on ext2), the results may prove interesting. Which is to say, should testing prove absolutely horrible results, proper disclaimers and warnings should be made readily available to avoid its use. Which is probably not a bad idea to begin with. ;) Nonetheless, it's an unknown right now in my mind. Hopefully some testing my reveal what reasonable expectations we should hold so that we can knowingly advise accordingly. Regards, -- Greg Copeland <greg@copelandconsulting.net> Copeland Computer Consulting
On Sat, Feb 01, 2003 at 11:30:17AM -0600, Greg Copeland wrote: > On Sat, 2003-02-01 at 00:34, Adam Haberlach wrote: > > On Sat, Feb 01, 2003 at 12:27:31AM -0600, Greg Copeland wrote: > > > On Fri, 2003-01-31 at 14:36, Dave Page wrote: > > > Please go with XFS or ext3. There are a number of blessed and horror > > > stories which still float around about reiserfs (recent and old; even > > > though I've never lost data with it -- using it now even). > > > > > > Might be worth testing FAT32 on NT as well. Even if we don't advocate > > > it's use, it may not hurt to at least get an understanding of what one > > > might reasonably expect from it. I'm betting there are people just > > > waiting to run with FAT32 in the Win32 world. ;) > > > > You'd better go with NTFS. There are a number of blessed and horror > > stories which still float around about FAT32 (recent and old; even though > > I've never lost data with it -- using it now even now. > > > > Might be worth testing reiserfs on Linux as well. Even if we don't > > advocate it's use, it may not hurt to at least get an understanding of > > what one my reasonably expect from it. I'm better there are people > > just waiting to run with reiserfs in the Linux world. ;) > > > > Regards, and tongue firmly in cheek, > > Touche! :P > > While I understand and even appreciate the humor value, I do believe the > picture is slightly different than your "analysis". If we make > something that runs on Win32 platforms, might it also run on Win98, > WinME, etc.? Let's face the facts that should it also run on these > platforms, it's probably only a matter of time before someone has it > running on FAT32 (even possible on NT, etc). In other words, I'm fully > expecting the lowest common denominator of MySQL user to be looking at > PostgreSQL on Win32. Which potentially means lots of FAT32 use. And > yes, even for a "production" environment. Ack! Double-ack! I was just trying to point out the inherent elitist bias in saying that Microsoft's old filesystem should be tested, even though it's use is discouraged, while one of Linux's new filesystems shouldn't, even though it's use is popular. There's a huge double standard here, caused by fear, uncertainty, and doubt. I'm just personally pretty tired of this anti-Microsoft bias. I'm going to be frank and say that many of the people here sound like a bunch of elitist assholes who refuse to sully themselves with a well-used, well-supported, and lately very useful operating system. For those of you who know my history (or care), I've supported Solaris, Linux, Win NT from 3.51 on up, and worked at one of the non-Linux "anti-Microsoft" companies for 4 years. I worked in a testlab where we tested and broke NT (usually caused by driver failures), SCO Unix (who KNOWs what made it panic), and Netware (effectively bulletproof). The fact is, the Win32 platform is mature. It is a modern operating system, and just because they don't do some things the way that the old guard Unix greybeards to doesn't necessarily make it inferior. What's more, all of this testing of filesystems is pretty moot at some point. So far, I've lost more data and time due to a bad locking procedure during vacuums (forcing a lot of quiet in-field upgrades from 7.2.2 to 7.2.3) then I ever expect to lose due to power failure. If we could spend a little more time testing the actual product and a little less time worrying about the underlying operating system, I'd be pretty happy. ...and I'm done discussing the issue for now, since there's not much more for me to say... -- Adam Haberlach | "If I have hacked deeper than them, it is adam@mediariffic.com | because I stand in their trenches." http://mediariffic.com | -- Graham Nelson
> -----Original Message----- > From: Christopher Kings-Lynne [mailto:chriskl@familyhealth.com.au] > Sent: 01 February 2003 12:40 > To: Greg Copeland > Cc: Dave Page; PostgresSQL Hackers Mailing List; Tom Lane > Subject: Re: [HACKERS] Win32 port powerfail testing > > > Try it with FreeBSD's UFS and FreeBSD 5.0's new UFS2 > filesystems perhaps - or I could! OK thanks for the comments everyone. Due to the fact that I really need to do quite a few tests and this might take a fair while, I'm going to limit this to ext3 and NTFS5. I'm all for the Win32 port, but if there's one thing I firmly believe it's that we should stongly recommend Windows 2000+ with NTFS as a minimum OS in all our docs. At a push I might agree to NT4 :-) As there have been no objections to my test program, I'll assume everyone agrees that it should test what is required and therefore expect not to hear 'but you didn't do...' if I end up in the same predicament as Hannu's friend!! Regards, Dave.