At 10:31 27-09-02 -0700, Jim Sibley wrote: >The best results I've gotten are by setting both values to zero, causing data >to written to disk almostly continously at almost disk speed rates.
It probably depends on the applications a lot. IMHO the benefit of waiting a while before writing dirty pages is not as much the efficiency of more work in one I/O, but the fact that you may avoid I/O completely. Consider a log file where lines of 64 bytes are appended each time. If you would write a dirty page immediately you write 64 times the same page, each with one more line. In particular benchmarks benefit a lot from this ;-) If I had time to measure the effect, I would decrease the maximum number of pages flushed out at once, and then probably decrease the percentage at which bdflush kicks in. Rob
