It seems that I've got to do something around packet layer to pass my exams. I've chosen packet-writing, so: what's the most important todo? --- INTERNATIONAL JOURNAL OF UNCERTAINTY FUZZINESS AND KNOWLEDGE-BASED SYSTEMS
Em Qui, 2003-06-19 às 17:30, O.Sezer escreveu:
Stabilizing it more seems as a more important TODO for me. We still have the "data corruption" issue unresolved yet.
Regards; O. Sezer
I completely agree. By the way, does the "data corruption" also occur in 2.5 kernels? Best, Paulo -- Paulo José da Silva e Silva Professor Assistente do Dep. de Ciência da Computação (Assistant Professor of the Computer Science Dept.) Universidade de São Paulo - Brazil e-mail: rsilva@ime.usp.br Web: http://www.ime.usp.br/~rsilva Teoria é o que não entendemos o (Theory is something we don't) suficiente para chamar de prática. (understand well enough to call practice)
On 19 Jun 2003, Paulo J. S. Silva wrote:
Em Qui, 2003-06-19 às 17:30, O.Sezer escreveu:
Stabilizing it more seems as a more important TODO for me. We still have the "data corruption" issue unresolved yet.
Regards; O. Sezer
I completely agree.
By the way, does the "data corruption" also occur in 2.5 kernels?
I have not been able to trigger data corruption with the 2.5.70 patch using an internal IDE CDRW. As far as I know, the corruption in 2.4 started with 2.4.10-pre11, which was when block device caching moved from the buffer cache to the page cache. My guess is that change somehow interferes with the packet patch. If that's true, the data corruption bug can not exist in the 2.5 driver, because that driver doesn't use the page/buffer caches at all. -- Peter Osterlund - petero2@telia.com http://w1.894.telia.com/~u89404340
Peter Osterlund wrote:
I have not been able to trigger data corruption with the 2.5.70 patch using an internal IDE CDRW. As far as I know, the corruption in 2.4 started with 2.4.10-pre11, which was when block device caching moved from the buffer cache to the page cache. My guess is that change somehow interferes with the packet patch. If that's true, the data corruption bug can not exist in the 2.5 driver, because that driver doesn't use the page/buffer caches at all.
How about 2.4-rmap then ?? Regards; Özkan Sezer
Peter Osterlund wrote:
I have not been able to trigger data corruption with the 2.5.70 patch using an internal IDE CDRW. As far as I know, the corruption in 2.4 started with 2.4.10-pre11, which was when block device caching moved from the buffer cache to the page cache. My guess is that change somehow interferes with the packet patch. If that's true, the data corruption bug can not exist in the 2.5 driver, because that driver doesn't use the page/buffer caches at all.
How about 2.4-rmap then ??
Sorry for the incomplete question: Is it due to rmap and / or bio in 2.5? (I'm completely ignorant of 2.5 changes..) Regards, Özkan Sezer
On Fri, 20 Jun 2003, O.Sezer wrote:
Peter Osterlund wrote:
I have not been able to trigger data corruption with the 2.5.70 patch using an internal IDE CDRW. As far as I know, the corruption in 2.4 started with 2.4.10-pre11, which was when block device caching moved from the buffer cache to the page cache. My guess is that change somehow interferes with the packet patch. If that's true, the data corruption bug can not exist in the 2.5 driver, because that driver doesn't use the page/buffer caches at all.
How about 2.4-rmap then ??
Sorry for the incomplete question: Is it due to rmap and / or bio in 2.5? (I'm completely ignorant of 2.5 changes..)
The buffer cache manipulation in the packet driver was replaced by bio stuff in 2.5, so my guess is that using 2.4-rmap will not make the corruption go away. -- Peter Osterlund - petero2@telia.com http://w1.894.telia.com/~u89404340
The buffer cache manipulation in the packet driver was replaced by bio stuff in 2.5, so my guess is that using 2.4-rmap will not make the corruption go away.
Bingo... I was curious to see if rmap would make a difference. I have then compiled a rmap kernel (I am running it right now). I can still get the corruption after copying my /usr/src/linux directory. Two side notes: 1) I have been playing on using ext2 + packet writing. It seems much faster than UDF on copying /usr/src/linux (many small files). Have anyone else the same experience? Is there any drawback to use ext2 but having cd-rw's that are only readable in linux? 2) I'll probably try 2.5 sometime soon. It is supposed to be much more responsive than 2.4, right? Usually, while I am writing to the cd-rw my machine becomes useless with 2.4. Hope this is fixed in 2.5. Have anyone managed to try 2.4 with Con Kolivas patch + packet writing? Is it more responsive than vanilla 2.4? Paulo -- Paulo José da Silva e Silva Professor Assistente do Dep. de Ciência da Computação (Assistant Professor of the Computer Science Dept.) Universidade de São Paulo - Brazil e-mail: rsilva@ime.usp.br Web: http://www.ime.usp.br/~rsilva Teoria é o que não entendemos o (Theory is something we don't) suficiente para chamar de prática. (understand well enough to call practice)
On 21 Jun 2003, Paulo J. S. Silva wrote:
2) I'll probably try 2.5 sometime soon. It is supposed to be much more responsive than 2.4, right? Usually, while I am writing to the cd-rw my machine becomes useless with 2.4. Hope this is fixed in 2.5.
It seemed much more responsive when I did some tests with 2.5.70 some time ago. I started a big copy to a slow (2x) CDRW on a 200MHz PPro with 96Mb RAM. Then I ran some tests. Starting big programs like emacs and netscape was quite ok, running a configure script was really slow but didn't stall completely. Running a compile job was noticably slower than usual, but didn't suffer as much as the configure script. I think that the more memory allocations a job requires, the more it will suffer in this scenario, because the copy operation will try to steal all free memory, which probably means that all memory allocations will have to wait for at least some data to be flushed to disk before they succeed. (With reservation that I don't really understand the internals of the linux virtual memory subsystem.) -- Peter Osterlund - petero2@telia.com http://w1.894.telia.com/~u89404340
1) I have been playing on using ext2 + packet writing. It seems much faster than UDF on copying /usr/src/linux (many small files). Have anyone else the same experience? Is there any drawback to use ext2 but having cd-rw's that are only readable in linux?
see: Re: mixed read/write operation From: Peter Osterlund (petero2_at_telia.com) Date: Thu Jul 18 2002 - 23:31:21 CEST
2) I'll probably try 2.5 sometime soon. It is supposed to be much more responsive than 2.4, right? Usually, while I am writing to the cd-rw my machine becomes useless with 2.4. Hope this is fixed in 2.5. Have anyone managed to try 2.4 with Con Kolivas patch + packet writing? Is it more responsive than vanilla 2.4?
It seem to have problems with the backport of O(1) scheduler. correct me if i wrong. Carlo
participants (6)
-
Carlo -
Maciek Fijalkowski -
Nicholas Wourms -
O.Sezer -
Paulo J. S. Silva -
Peter Osterlund