Re: speed of -CURRENT [was: questions about the state of current]

[Stijn Hoop]

On Wed, Oct 30, 2002 at 07:48:14AM -0500, Alexander Kabaev wrote:
> > I am experiencing a really noticable slower startup time on my very
> > recent-CURRENT laptop for almost all programs. The problem seems to be
> > in getting info in the cache, because it disappears when I start the
> > same program again.
> 
> This almost certainly is caused by the 'ioslow' addition to
> specfs_vnops.c. Find a block in specfs_strategy function which goes into
> tsleep for niced processes and comment it out. Let us know if that helps
> :)

Yes, that's it. -CURRENT actually feels snappier than -STABLE now :)

Below is the diff that I used. Will something other than I/O for
niced processes break using this?

Thanks!

--Stijn

--- spec_vnops.c.orig   Mon Oct 28 08:07:49 2002
+++ spec_vnops.c        Wed Oct 30 14:22:01 2002
@@ -530,17 +530,19 @@
        struct mount *mp;
        int error;
        struct cdevsw *dsw;
-       struct thread *td = curthread;
+/*     struct thread *td = curthread; */
        
        /*
         * Slow down disk requests for niced processes.
         */
+/* XXX: per Alexander Kabaev mail 2002/10/30 07:48 -5
        if (td && td->td_ksegrp->kg_nice > 0) {
                mtx_lock(&strategy_mtx);
                msleep(&strategy_mtx, &strategy_mtx,
                    PPAUSE | PCATCH | PDROP, "ioslow",
                    td->td_ksegrp->kg_nice);
        }
+*/
        bp = ap->a_bp;
        vp = ap->a_vp;
        if (bp->b_iocmd == BIO_WRITE) {

msg45645/pgp00000.pgp
Description: PGP signature