Re: reiserfsck --rebuild-tree is very slow

[Hans Reiser <reiser@namesys.com>]

Very interesting.  Probably we should warn users about that when they 
run fsck.  Vitaly?

Hasns

berthiaume_wayne@emc.com wrote:

>	You may want to check with hdparm -d /dev/hd<x> to see if the drive
>has dropped out of DMA into PIO. When IDE detects an error it will step back
>and retry in PIO, afterwards it remains in PIO. In 2.4.19 there is a kernel
>config parameter to set IDE to always DMA so you don't get this type of
>system degradation.
>Wayne.
>-----Original Message-----
>From: Oleg Drokin [mailto:green@namesys.com]
>Sent: Monday, August 05, 2002 3:43 AM
>To: Patrik Wallstrom
>Cc: reiserfs-list@namesys.com
>Subject: Re: [reiserfs-list] reiserfsck --rebuild-tree is very slow
>
>
>Hello!
>
>On Mon, Aug 05, 2002 at 09:37:52AM +0200, Patrik Wallstrom wrote:
>  
>
>>>>The result of reiserfsck --check and the result with --fix-fixable:
>>>>bread: Cannot read a block # 10052279.
>>>>        
>>>>
>>>Hm, what's in kernel logs, is it filled with I/O errors?
>>>      
>>>
>>Aug  5 06:34:13 motor kernel: hdd: lost interrupt
>>Aug  5 06:34:53 motor last message repeated 4 times
>>Aug  5 06:35:53 motor last message repeated 6 times
>>    
>>
>
>Hm, this is a problem of IDE layer then
>
>  
>
>>>>From what I can see, this session can last for many years...
>>>>Any advise?
>>>>        
>>>>
>>>Look into the kernel logs, if it is filled by many I/O error messages, 
>>>then it will indeed may atek a lot of time. Also it is possible that the
>>>drive itself is reading data very slowly (ie it encounters multiple
>>>      
>>>
>errors
>  
>
>>>that it tries to correct and this takes huge amount of time, I saw such
>>>behaviour with IBM DTLA drives).
>>>If you can access data on the drive just fine without fsck
>>>(I mean if dd if=/dev/device of=/dev/zero bs=4096k works as fast as
>>>      
>>>
>expected),
>  
>
>>>then this may need more investigations.
>>>      
>>>
>>The read operation from dd was very fast.
>>    
>>
>
>You mean right now if you interrupt reiserfsck and issue a dd command, it
>will be fast (ie. you tried it right now), or it was fast some time ago?
>
>How aboult changing blocksize from 4M down to 4k?
>
>  
>
>>Should I just try --rebuild-tree, and wait for some weeks? :)
>>    
>>
>
>No, you should fix a problem in your IDE subsystem instead.
>
>  
>
>>The error from --check was now
>>Bad root block 4294967295. (--rebuild-tree did not complete)
>>    
>>
>
>Yes, this is correct message for unfinished rebuild tree.
>
>  
>
>>So I guess I have to...
>>    
>>
>
>As an alternative thing: you can copy your partition to file, run fsck
>on that file and then dd the file back, but this way the problem you
>are having won't be fixed. And we have nothing to do with the problem so
>we cannot help here (err, we can help probably but since that's not reiserfs
>bug, such a help won't be free).
>
>Bye,
>    Oleg
>
>
>  
>


-- 
Hans