On May 20, 2007, at 8:50 AM, Michael Bartl wrote:
But the good thing is, that (even without a reason) the problems were
gone when we started to use raw devices instead of filesystems. ...
Makes one wonder if excessive, delayed file system sync operations
were the issue, in writes. This might be tested by performing pure
read operations in the file system ('cp SomeFile /dev/null' is a
basic test, versus copying file to file).
Richard Sims
|