[U-Boot] [PATCH] common/cmd_ext_common: measure throughput

[Tom Rini]

On Wed, Oct 17, 2012 at 12:16:14PM +0200, Andreas Bie??mann wrote:
> Dear Wolfgang Denk,
> 
> On 17.10.2012 12:05, Wolfgang Denk wrote:
> > Dear Andreas Bie??mann,
> > 
> > In message <1350467910-2014-1-git-send-email-andreas.devel at googlemail.com> you wrote:
> >> This patch adds time measurement and throughput calculation for the ext2load and
> >> ext4load commands.
> > ...
> >> +	unsigned long time_start;
> > ...
> >> +	time_start = get_timer(0);
> >>  	if (ext4fs_read((char *)addr, filelen) != filelen) {
> >>  		printf("** Unable to read \"%s\" from %s %d:%d **\n",
> >>  		       filename, argv[1], dev, part);
> >>  		ext4fs_close();
> >>  		goto fail;
> >>  	}
> >> +	time_start = get_timer(time_start);
> > 
> > There, "time_start" is clearly a mis-nomer.  How about
> > s/time_start/time/ ?
> 
> sounds better, however this is a plane copy from Simons tftp measurement
> patch.
> 
> >> +		print_size(filelen / time_start * 1000, "/s");
> > 
> > Does this give reasonable results for small files, say when loading a
> > 20 byte file ?
> 
> Well, possible no:
> 
> ---8<---
> U-Boot> ext2load mmc 0 10020000 /etc/hosts
> Loading file "/etc/hosts" from mmc device 0:1
> 20 bytes read in 0 ms
> U-Boot> ext2load mmc 0 10020000 /etc/shadow
> Loading file "/etc/shadow" from mmc device 0:1
> 95 bytes read in 0 ms
> U-Boot>  ext2load mmc 0 10020000 /etc/passwd
> Loading file "/etc/passwd" from mmc device 0:1
> 366 bytes read in 0 ms
> U-Boot> ext2load mmc 0 10020000 /etc/services
> Loading file "/etc/services" from mmc device 0:1
> 18465 bytes read in 3 ms (5.9 MiB/s)
> U-Boot>
> --->8---
> 
> But as you see extremely short transfers are omitted due to time
> difference of '0' (at least on my avr32 system here).
> The main aim for this patch was to measure performance gain of Josh Wu's
> gen_atmel_mci patch for multiple block access, hopefully this is useful
> for others.
> I would like to have some feedback how the measurement is for very small
> files on other systems. Then I could provide a v2 which uses another
> variable name for the time.

I'm fine with not giving a speed on <1 ms transactions.  Lets see a v2
with the new variable name, thanks!

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20121025/483996ba/attachment.pgp>