civileme wrote: > Guy Zelck wrote: > >> Lyvim Xaphir wrote: >> >>> On Thu, 2002-03-28 at 07:26, Guy Zelck wrote: >>> >>> ---snip--- >>> >>>> Interesting thread I thought to post my troubles with XFS to. >>>> When I installed md8.1 on my home system I went from reiserfs, >>>> which never gave me trouble, to using xfs which I knew from work >>>> having it on our Silicon Graphics machine. I also read a lot about >>>> it and for speed and features it seemed the best and having a high >>>> esteem for SGI I did not hesitate. >>>> >>>> Nevertheless I've experienced about 4 times so far the following (I >>>> have an IDE system disk 'quantum fireball' of 8GB): >>>> Every time there's a power cut I find a lot of files corrupted. >>>> Instead of their original contents they contain nothing but ^@ >>>> (viewed with vim) characters. The files concerned are e.g. all the >>>> KDE config files of apps that were open at the time of the power >>>> cut, but I once had inittab & message affected too. You can imagine >>>> that KDE wouldn't start after that in the 1st case and the whole >>>> system was fucked in the 2nd. This for me >>>> is totally unacceptable and I don't dare to imagine what this would >>>> mean for a company's production machine. >>>> >>> >>> Your results are interesting. But as in most other things it's >>> important to have a reference point(s) for comparison. So I've got some >>> questions: >>> >>> 1) What is the history of this hardware with relation to the previously >>> installed scheme(s)? (what filesystem, Linux version, Win version) >>> >> The history is very recent since I bought this machine in Nov 2001 >> but lets start with my previous setup, this was : >> An AbitBX6 mb with a PII 266MHz, Intel BX440 chipset, 384 MB memory >> (100Mhz sdramm) and a completely SCSI setup based on the Adaptec >> AHA-2940 with Seagates of 512MB, 1GB & 2GB (3x) (old disks <=1995). >> My previous os was md7.2 and I used reiserfs for everything exept >> /boot (ext2). I suffered block ups and powerfails like every non >> ups'ed person but never had corruptions. >> >> My present new machine is from Medion, sold by Aldi and has a 1.8GHz >> P4 on a msi mb (I'm not totally sure here, the manual is made for >> complete newbies and they avoid speaking about it alltogether), the >> chipset is Intel i845, there was 256MB (130MHz sdramm) memory on >> board by I added another 256MB, it has a Seagate st380020A (80GB) >> which has XP Home edition installed on it by Medion but I added a >> Quantum Fireball of 8GB for Linux (The 2 os'es are completely >> separated). The disks are on separate ide channels too with a dvd and >> cdrw as slave respectively. I use XFS for everything except /boot >> (ext2). I run md8.1 on it, stock kernel 2.4.8. >> I incorporated my complete SCSI chain from my previous machine to >> this new setup and use this as extra disk space for storage purposes >> and it still has my md7.2 on it as a backup. >> >>> >>> >>> 2) Can you recall crashes under other circumstances that did not >>> involve >>> XFS in any way? (under this hardware config) Focus a little on hard >>> drive history, as well as everything else. >>> >> I had a blockup which was os related one time and in that >> circumstance I had no corruption I think, but I was just lucky. But >> listen to the next: after having migrated my scsi chain I had one >> scsi disk that began to fall apart (from 1995 this one) and by doing >> so blocked my scsi bus which blocked the whole machine. I had to >> reboot and there I discovered to my horror and discomfort the >> corruptions for the 1st time. You can recognize the files as 'ASCII >> text, with no line terminators' with cmd 'file'. I had this 2 more >> times until this disk was 100% knock-out. I finally took it out and >> had no corruptions anymore. I more or less blamed reiserfs at 1st as >> I thought its interaction with the kernel blocked the working of xfs >> somehow but this makes no sense, even if xfs is blocked it should >> recover afterwards. Now, the 25 of March I suffered a plain power >> failure and bingo, the same corruptions appeared again. So now I'm >> sure it has nothing to do with my SCSI chain or reiserfs. >> >>> >>> 3) What is the hardware's history with respect to diagnostics and >>> tests? Has the memory been checked, dos diagnostics run, cpu checked.. >>> >> Never ran any tests, have no experience with this, and since all my >> stuff is brand new... I never had any indication that sth. would be >> wrong with my cpu or memory. There is one thing I'm puzzled about and >> that is that my AHA-2940 scsi card's BIOS doesn't allow me to >> configure or diagnose my scsi disks anymore since it was moved to >> this new machine. PCI bus incompatibilities? But on bootup it still >> nicely recognizes all scsi disks and starts them one by one. In using >> the disks I have no problems as I use these scsi disk to do my >> downloads too. >> >>> >>> >>> If not, has the hardware got a respectable history of problem free >>> operation under the previous OS? yay or nay...plus some details >>> would be >>> nice. >>> >> As you could read from the previous this hw dates only from Nov 2001 >> and I never had any abnormalities with this new h.w.. The troubles >> came from my legacy scsi hw I moved over but after the faulty scsi >> disk was singled out and removed I ran quite a long time problem free >> (from the beginning of this year). >> >> Thanks for taking serious intrest in this matter. >> I also thank Civileme for posting his experiences. >> >> Guy. >> >> >> >> ------------------------------------------------------------------------ >> >> Want to buy your Pack or Services from MandrakeSoft? Go to >> http://www.mandrakestore.com >> > Yep, thanks for asking, Lyvim, and for answering, Guy. > > NEW is something that should always be checked. Infant mortality in > electronic devices is common and failures are rare from infancy on to > about 5 years or so... KInd of an interesting switch on the usual > Poisson distribution, but nevertheless true. > > memtest-x86.bin is on your CD1 in the images directory and can be sent > to floppy with a dd and the floppy can be booted and run to test > memory. An initial 512 Mb may not show problems for quite a while, > even running linux. I had the unpleasant experience of negotiating a > warranty return on a 512M DDR recently, and my system would run for > days then suddenly reset without any indication in the logs of any > temperature conditions.... > > Civileme > I've installed it and ran it for 6H without any errors popping up. I personally don't believe it's the memory which is in fault. As if the devil was involved I had a freeze again just after boot-up. I followed the Alt-SysRq-... sequence to finally reboot but then again I had my KDE files corrupted. What I don't understand is how that crap is written to my config files. If xfs writes what it finds in its xfs log then it would mean that the xfs log contains rubbish. How did this rubbish get in there or is it taking a wrong part of the log.
Is the xfs that's used in md8.2 still the same version or is it a higher version? Maybe upgrade? When I boot up the only fsck I ever notice amongst Aurora's output is the one of my /boot partition (ext2), is this supposed to be like that? I suppose Fsck in rc.sysinit just re-directs the output to syslog. Guy.
Want to buy your Pack or Services from MandrakeSoft? Go to http://www.mandrakestore.com
