Re: 2.6.12-rc2 and as-iosched

Re: 2.6.12-rc2 and as-iosched

[Kenneth Parrish]

Randy> Need more info.

        Greetings.  :)
CONFIG_HZ_ changes the block device elevator time-out values -- didn't see.

Re: 2.6.12-rc2 and as-iosched

[Jens Axboe]

On Mon, Jul 18 2005, Kenneth Parrish wrote:
> Randy> Need more info.
> 
>         Greetings.  :)
> CONFIG_HZ_ changes the block device elevator time-out values -- didn't see.

I cannot reproduce here with cfq and HZ == 250, the jiffies <-> msec
conversions are working fine. Please provide a proper bug report, did
you change the values and not getting the expected back, or what is
going wrong??

-- 
Jens Axboe

Re: 2.6.12-rc2 and as-iosched

[Jens Axboe]

On Mon, Jul 18 2005, Jens Axboe wrote:
> On Mon, Jul 18 2005, Kenneth Parrish wrote:
> > Randy> Need more info.
> > 
> >         Greetings.  :)
> > CONFIG_HZ_ changes the block device elevator time-out values -- didn't see.
> 
> I cannot reproduce here with cfq and HZ == 250, the jiffies <-> msec
> conversions are working fine. Please provide a proper bug report, did
> you change the values and not getting the expected back, or what is
> going wrong??

ok, AS is definitely broken, it does an internal HZ <-> msec conversion
in the store/show functions as well. This should fix it.

--- /opt/kernel/linux-2.6.13-rc3/drivers/block/as-iosched.c	2005-07-13 06:46:46.000000000 +0200
+++ linux-2.6.13-rc3/drivers/block/as-iosched.c	2005-07-18 15:46:23.000000000 +0200
@@ -1935,23 +1935,15 @@
 static ssize_t
 as_var_show(unsigned int var, char *page)
 {
-	var = (var * 1000) / HZ;
 	return sprintf(page, "%d\n", var);
 }
 
 static ssize_t
 as_var_store(unsigned long *var, const char *page, size_t count)
 {
-	unsigned long tmp;
 	char *p = (char *) page;
 
-	tmp = simple_strtoul(p, &p, 10);
-	if (tmp != 0) {
-		tmp = (tmp * HZ) / 1000;
-		if (tmp == 0)
-			tmp = 1;
-	}
-	*var = tmp;
+	*var = simple_strtoul(p, &p, 10);
 	return count;
 }
 

-- 
Jens Axboe

Re: 2.6.12-rc2 and as-iosched

[Kenneth Parrish]

-=> axboe@suse.de wrote to Kenneth Parrish <=-

 ax> ok, AS is definitely broken, it does an internal HZ <-> msec
 ax> conversion in the store/show functions as well. This should fix
 ax> it.
thank ya   :-)

2.6.12-rc2 and as-iosched

[Kenneth Parrish]

========================================================================
 System: FamilyNet HQ
   Area: Internet-Email, Internet E-mail
   Date: Jul 16 2005  22:45
   From: Kenneth Parrish
     To: linux-kernel@vger.kernel.org
   Subj: 2.6.12-rc2 and as-iosched
------------------------------------------------------------------------
v2.6.13-rc2 as-iosched.c and /sys/block/hda/queue/iosched/* values differ:
[..]
Oops, HZ=250, so correct.

... The later PDP-11s supported 4MB.
--- MultiMail/Linux v0.46

Re: 2.6.12-rc2 and as-iosched

[randy_dunlap]

On 17 Jul 2005 06:06:08 GMT Kenneth Parrish wrote:

> ========================================================================
>  System: FamilyNet HQ
>    Area: Internet-Email, Internet E-mail
>    Date: Jul 16 2005  22:45
>    From: Kenneth Parrish
>      To: linux-kernel@vger.kernel.org
>    Subj: 2.6.12-rc2 and as-iosched
> ------------------------------------------------------------------------
> v2.6.13-rc2 as-iosched.c and /sys/block/hda/queue/iosched/* values differ:
> [..]
> Oops, HZ=250, so correct.

Hi,

My ESP isn't working so well tonight.

What are you trying to report here?
Need more info.

---
~Randy

2.6.12-rc2 and as-iosched

[Kenneth Parrish]

v2.6.13-rc2 as-iosched.c and /sys/block/hda/queue/iosched/* values differ:

/*  from ../drivers/block/as-iosched.c  */
#define default_read_expire (HZ / 8)
#define default_write_expire (HZ / 4)
#define default_read_batch_expire (HZ / 2)
#define default_write_batch_expire (HZ / 8)

$ ls /sys/block/hda/queue/iosched
antic_expire
est_time
read_batch_expire
read_expire
write_batch_expire
write_expire

$ cat /sys/block/hda/queue/iosched
16
7 % exit probability
0 ms new thinktime
647336 sectors new seek distance
2000  # read_batch_expire (500ms?)
496   # read_expire (125ms?)
496   # write_batch_expire (125ms?)
992   # write_expire (250ms?)

--