Bram Stolk wrote: >On Mon, 21 Jan 2002 16:21:30 +0300 >Nikita Danilov <[EMAIL PROTECTED]> wrote: > >>Bram Stolk writes: >> > Hello, >> > >> > I'm trying to delay the writes to disk. >> > >> > Normally, this can be achieved by editting >> > /proc/sys/vm/bdflush >> > >> > I believed reiserfs does not use bdflush nor kupdated. >> > Looking at the source code, I get the impression that >> > the write delay is hard coded to 5 seconds. >> > Is this so? >> > >> > The line of code that makes me think it is hardcoded >> > to 5 is this line: >> > >> > interruptible_sleep_on_timeout(&reiserfs_commit_thread_wait, 5) ; >> >>Upgrade to the newer kernel, by the way: code line above contains bug >>already fixed (interruptible_sleep_on_timeout() takes timeout in jiffies >>rather than second). Back to the your question, there is no currently >>way to tune this delay. I thought about reusing /proc/sys/vm/bdflush, >>but this thing is not available to modules, as Chris Mason pointed >>out. Probably our own parameter in /proc, or boot/module parameter is >>the right way to go. >> >> > >> > Is there an easy way to have the filesystem delay the writes much longer, >> > like an hour or so? >> >>You will risk loosing a lot of data then, but if you want to do this, >> > >losing data is something I can live with. >But will less-frequent writing also increase chances of leaving >the filesystem in an invalid state? Lost data with an otherwise >valid fs is not so bad. > >>just edit the code. It would be nice if you measured how this impacts >>performance. >> > >Ok, a code-edit. >But is it indeed the '5' value in the interruptible_sleep_on_timeout call? > >The latest 2.4.* kernel on ftp.funet.fi (2.4.18) contains the same line, >also with value '5'. What 'newer kernel' are you talking about? the 2.5 kernel? > >About performance: it will undoubtedly improve a lot if you only seldom >burden the disk with write-ops. It should especially benefit from cases >where you repeatedly write the same disk-blocks. > > Bram > >> > >> > many thanks, >> > >> > Bram Stolk >> > >> >>Nikita. >> > > I thought that I requested that meta-data and data be given the same delay (30 seconds)? What happened to my request?
Oleg, track and resolve this issue please. Hans
