Re: Suggested code change : Simple : Scale pdflush threads from desktop to server

[Chris Snook <chris.snook@gmail.com>]

On Fri, Jul 10, 2009 at 9:25 PM, Mitchell Erblich<erblichs@earthlink.net> wrote:
> Group,
>
>                pdflush threads clean dirty pages
>
>          Under the past simple assumption that a greater number of
>          page daemon threads will have the TENDENCY to clean
>          the pages faster.
>
>        Another assumption is that a server will have at least 2x / 4x the
>        number of drives and memory, so allocating more pdflush() threads
>        makes sense.

That's a rather sweeping generalization.

>        Relying on a recent change, code base on whether the system is
>        a desktop or a server, scale the number of pdthreads() which would
>        result in the below code change.
>
>        The suggestion is to double the MIN number of threads and set the
>        MAX number to 4x.
>
>        ./mm/pdflush.c
>        /* Scale for a server */
>        #define MIN_PDFLUSH_THREADS     4               /* 2x desktop value
> */
>        #define MAX_PDFLUSH_THREADS     32              /* 4x desktop value
> */

So, you're taking a well-established and empirically validated set of
constants, and changing it only in the case where users are least
tolerant of change?  I agree that the existing pdflush tuning is a bit
of a kludge, but this just adds more noise to the data we need to
analyze and optimize pdflush.

>        /*
>         * secondary suggestion is to add a DEBUG type /var/log/system
> messages that
>         * will rate limit independent of desktop or server.
>         */
>
>        else if (nr_pdflush_threads == MAX_PDFLUSH_THREADS) {
>                   /* optional PDFLUSH msg */
>                   if (printk_ratelimit() {
>                                                    printk(KERN_INFO
> "MAX_PDFLUSH_THREADS Limited\n");
>                   }
>         }

Log messages, even at debug level, for normal conditions are a really bad idea.

pdflush definitely needs work, but twiddling constants will, at best,
optimize it for a small subset of the user base for a brief period of
time.  We really need patches that do intelligent things based on
system resources and load, and we need data to support them.

-- Chris