Re: Linux-2.4.1-pre11 / ll_rw_b watermark metric?

Re: Linux-2.4.1-pre11 / ll_rw_b watermark metric?

[Dieter Nützel]

I have pre11 running with Andrea's suggested fix.

        high_queued_sectors = total_ram / 3;
        low_queued_sectors = high_queued_sectors / 2;
        if (low_queued_sectors < 0)
                low_queued_sectors = total_ram / 2;
 
        /*
         * for big RAM machines (>= 384MB), use more for I/O
         */
        /*
        if (total_ram >= MB(384)) {
                high_queued_sectors = (total_ram * 4) / 5;
                low_queued_sectors = high_queued_sectors - MB(128);
        }
        */

Shouldn't it be clean for a 2.4.1 release?

-Dieter
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

Re: Linux-2.4.1-pre11 / ll_rw_b watermark metric?

[Jens Axboe]

On Mon, Jan 29 2001, Dieter Nützel wrote:
> I have pre11 running with Andrea's suggested fix.
> 
>         high_queued_sectors = total_ram / 3;
>         low_queued_sectors = high_queued_sectors / 2;
>         if (low_queued_sectors < 0)
>                 low_queued_sectors = total_ram / 2;
>  
>         /*
>          * for big RAM machines (>= 384MB), use more for I/O
>          */
>         /*
>         if (total_ram >= MB(384)) {
>                 high_queued_sectors = (total_ram * 4) / 5;
>                 low_queued_sectors = high_queued_sectors - MB(128);
>         }
>         */
> 
> Shouldn't it be clean for a 2.4.1 release?

With enough swap the numbers I saw were not conclusive. I promised
to test which I haven't gotten done yet, I will do this tomorrow
and make sure we have the right ratios. However, I don't think
the pre11 numbers are much off - do you have any results?

-- 
Jens Axboe

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

Re: Linux-2.4.1-pre11 / ll_rw_b watermark metric?

[Dieter Nützel]

Am Montag, 29. Januar 2001 04:46 schrieb Jens Axboe:
> On Mon, Jan 29 2001, Dieter Nützel wrote:
> > I have pre11 running with Andrea's suggested fix.
> >
> >         high_queued_sectors = total_ram / 3;
> >         low_queued_sectors = high_queued_sectors / 2;
> >         if (low_queued_sectors < 0)
> >                 low_queued_sectors = total_ram / 2;
> >
> >         /*
> >          * for big RAM machines (>= 384MB), use more for I/O
> >          */
> >         /*
> >         if (total_ram >= MB(384)) {
> >                 high_queued_sectors = (total_ram * 4) / 5;
> >                 low_queued_sectors = high_queued_sectors - MB(128);
> >         }
> >         */
> >
> > Shouldn't it be clean for a 2.4.1 release?
>
> With enough swap the numbers I saw were not conclusive. I promised
> to test which I haven't gotten done yet, I will do this tomorrow
> and make sure we have the right ratios. However, I don't think
> the pre11 numbers are much off - do you have any results?

I have 256 MB RAM and 200 MB swap but nothing of the later was used during 
"dbench 48". It was nothing spectacular but a litte bit faster with the above.
Attention: Results only from memory...;-)

Good night.
	Dieter
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

Re: Linux-2.4.1-pre11 / ll_rw_b watermark metric?

[Petr Vandrovec]

On 29 Jan 01 at 4:43, Dieter Nützel wrote:
> I have pre11 running with Andrea's suggested fix.
> 
>         high_queued_sectors = total_ram / 3;
>         low_queued_sectors = high_queued_sectors / 2;
>         if (low_queued_sectors < 0)
>                 low_queued_sectors = total_ram / 2;
>  

I have one question: How it can happen that low_queued_sectors
is less than zero with this changed logic? (And if it get triggered, 
low_queued_sectors will be greater than high_queued_sectors - which 
is not what we want...)

But it is certainly better than 2.4.0-pre8 approach, as
with 200MB of memory (exactly 192MB left unused) you can end up with 
low_queued_sectors == 0... And it does not give you optimal behavior.
                                        Best regards,
                                            Petr Vandrovec
                                            vandrove@vc.cvut.cz
                                            
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

Re: Linux-2.4.1-pre11 / ll_rw_b watermark metric?

[Jens Axboe]

On Mon, Jan 29 2001, Petr Vandrovec wrote:
> On 29 Jan 01 at 4:43, Dieter Nützel wrote:
> > I have pre11 running with Andrea's suggested fix.
> >
> >         high_queued_sectors = total_ram / 3;
> >         low_queued_sectors = high_queued_sectors / 2;
> >         if (low_queued_sectors < 0)
> >                 low_queued_sectors = total_ram / 2;
> >
> 
> I have one question: How it can happen that low_queued_sectors
> is less than zero with this changed logic? (And if it get triggered,
> low_queued_sectors will be greater than high_queued_sectors - which
> is not what we want...)

This wasn't my change, but you're right it's wrong.

> But it is certainly better than 2.4.0-pre8 approach, as
> with 200MB of memory (exactly 192MB left unused) you can end up with
> low_queued_sectors == 0... And it does not give you optimal behavior.

Same here, definitely not right either. Dunno how I missed that.

-- 
Jens Axboe

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/