Hello Jacques, I reply here below:
Le Tuesday 28 February 2006 00:56, jacques Normand a écrit : > On Mon, Feb 27, 2006 at 11:33:19PM +0100, Juan Piñeros wrote: > > 1 Raw_Read_Error_Rate 0x000d 100 100 050 > > Pre-fail Offline > > - 51 > > 195 Hardware_ECC_Recovered 0x001a 100 100 000 > > Old_age Always > > - 2 > > 199 UDMA_CRC_Error_Count 0x003e 200 200 000 > > Old_age Always > > - 9 > > This is where your issue seems to live. I have never seen the read > error and ecc corrected number not matching. It would mean that an error > occurs but there has been no way to make it right so I would expect the > read to be garbage... Did you see any corruption in your files? I mean > data corrupted instead of metadata? * IN MACHINE1: In machine1, the files that were recovered after the crash were converted to 8 characters dos names, but the files I tried to open were ok (I can not open all of them since this is 10GB). There were some unrecoverable files (following the recover soft said) but I suppose these were previously deleted files that were partially overwritten before the crash. I never saw a file simply corrupted but still existing in machine1: simply they were ok, then suddenly a whole directory was lost. One thing I remember now is that I installed smartools in machine1 two months before the crash (the disk was 2 years in use without problems before), but this is maybe unrelated with the crash. * IN MACHINE2: Here, the most recent files that were lost and after that recovered were totally corrupted. However we avoided to write on the disk with linux, and the disk recovery function of windows (the "System Volume Information") was disabled. > > Also, you say that sata does not support smart. That is not true, with > one of the very recent kernels (2.6.15.4), you can get them. I have not > much experience with the kernels shipped with debian. I always recompile > my own. But some problems I had with an nfs server (in an HPC system) > vanished when I upgraded from 2.6.12 to 2.6.14. There was a bug with the > futex, and I think that was the source of my problems (race conditions > are always nasty). I can try to install a more recent kernel to make smart working, but the disk is new, is it useful? I have another question: why did etch installer chose the 2.6.12-1-386 kernel? Why not the 686? Maybe installing 686 can help? However this problem in machine2 remains classified as a bug and not as a problem related to user choices? > > As for the udma crc? That usually means that your controller/cable is > going bad. Each time I have seen that, the whole system crashed > corrupting files everywhere... That is pretty odd that you see the thig > on two different system though. So it seems that I have two different problems in the two machines? It seems that machine1's disk is to be replaced? > > jacques > > PS: With development kernels, always try to use the latest. Especially > when you see a problem. (And I still consider the 2.6 as being a > development version) So if you install a very recent kernel from kernel.org, you have to apply a lot of patches before having it working? Is it better that I install a stock 2.4 debian kernel on machine2? It can be difficult to downgrade from 2.6 to 2.4 (udev, etc). Thanks, Juan.