On Mon, 2003-09-08 at 11:08, Erwin Burgstaller wrote: > I actually found out by writing a 700MB file to a file system with 1GB > of size, removing it and again copying that file into that file system, > where copy then aborted with only 300 MB written. But one could do that > on small file systems (e.g. /boot) very fast and just triggering by time > could fail then too.
Interesting. Maybe I could set a flag if there are pending deletes, and any new allocation attempts would force the journal buffer to disk if that flag is set. We used to start I/O on partial journal pages any time there was no active I/O to the journal, but we saw a significant performance gain when we deferred those writes until the page was full, or at least until a synchronous transaction forced us to write it. I'd like to fix this problem without introducing a lot of unnecessary I/O. Thanks, Shaggy > > Erwin -- David Kleikamp IBM Linux Technology Center _______________________________________________ Jfs-discussion mailing list [EMAIL PROTECTED] http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion
