[U-Boot] USB storage performance EHCI question

[U-Boot] USB storage performance EHCI question

[Aaron Williams]

I just got EHCI support working with our chip and I was wondering what sort of 
performance I could expect using ext2?  Right now it seems extremely slow 
coming off of a fast solid-state drive. I'm getting around 100KB/second. It's 
running at full-speed (480Mbps). Is this normal?

We are basically using the standard EHCI driver with only a few minor 
modifications.

-Aaron

[U-Boot] USB storage performance EHCI question

[Wolfgang Denk]

Dear Aaron Williams,

In message <201101121955.04367.Aaron.Williams@caviumnetworks.com> you wrote:
> I just got EHCI support working with our chip and I was wondering what sort of 
> performance I could expect using ext2?  Right now it seems extremely slow 
> coming off of a fast solid-state drive. I'm getting around 100KB/second. It's 
> running at full-speed (480Mbps). Is this normal?

As you don't provide any of the important information let me guess:
this is on an ARM board, and you are using U-Boot version v2010.09 or
older?  If this is true then yes, this is to be expected.

Hm... I just notice the "caviumnetworks" in your address, so this is
probably MIPS?  Check if your caches are turned on...

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
The universe, they said, depended for its operation on the balance of
four forces which they identified as charm,  persuasion,  uncertainty
and bloody-mindedness.      -- Terry Pratchett, "The Light Fantastic"

[U-Boot] USB storage performance EHCI question

[Sergei Shtylyov]

Hello.

On 13-01-2011 9:09, Wolfgang Denk wrote:

> Hm... I just notice the "caviumnetworks" in your address, so this is
> probably MIPS?  Check if your caches are turned on...

    BTW, Cavium also produces ARMs (like CNS3xxx)...

> Best regards,

> Wolfgang Denk

WBR, Sergei

[U-Boot] USB storage performance EHCI question

[Aaron Williams]

On 1/14/2011 1:19 AM, Wolfgang Denk wrote:
> Dear Aaron Williams,
>
> please always keep the ML on cc:
>
> In message<4D2FA7F0.7020900@caviumnetworks.com>  you wrote:
>> It is based off of 2010.09 but I have applied all of the latest USB
>> updates from GIT with the exception of the multiple LUN support (this
>> breaks the SanDisk Cruzer because the Cruzer reports max lun=1). All
>> caches are enabled and this is on a 1.2GHz MIPS which is cache-coherent
>> (no need to flush/invalidate).
> Maybe there are other issues with your code.  It's hard to comment on
> onknown out-of-tree ports.
>
> Best regards,
>
> Wolfgang Denk
>
I don't think so. The only changes I made were to the register access 
macros to use a couple assembly functions since our EHCI registers 
require 64-bit addressing and a very quick virtual to physical 
conversion for the descriptors which is quite fast. Other than that, the 
code is stock from GIT. Note that we do not use interrupts either but 
polling.

Unless I'm reading it wrong it also looks like in EHCI it only uses one 
descriptor and I wonder if this might have something to do with it?

I can provide a diff when I return to the office next week.

-Aaron

[U-Boot] USB storage performance EHCI question - ext2fs performance

[Aaron Williams]

I  now have some more information on the performance issue. It appears to be 
filesystem related. If I format the drive as FAT32 I am getting a transfer 
rate of approximately 10MB/second. However, if I format it as ext3 I am seeing 
a transfer rate of only  101KB/second!  The only difference is the filesystem 
and both are clean filesystems.

I formatted the drive with:

mkfs.ext3 -j -L iomega -O dir_index,has_journal,large_file -t ext3 -v 
/dev/sdc1

mke2fs 1.41.11 (14-Mar-2010)
fs_types for mke2fs.conf resolution: 'ext3', 'default'
Calling BLKDISCARD from 0 to 1014095872 failed.
Filesystem label=iomega
OS type: Linux
Block size=4096 (log=2)
Fragment size=4096 (log=2)
Stride=0 blocks, Stripe width=0 blocks
61952 inodes, 247582 blocks
12379 blocks (5.00%) reserved for the super user
First data block=0
Maximum filesystem blocks=255852544
8 block groups
32768 blocks per group, 32768 fragments per group
7744 inodes per group
Superblock backups stored on blocks: 
        32768, 98304, 163840, 229376

Writing inode tables: done                            
Creating journal (4096 blocks): done
Writing superblocks and filesystem accounting information: done

This filesystem will be automatically checked every 35 mounts or
180 days, whichever comes first.  Use tune2fs -c or -i to override.

-Aaron

On Friday, January 14, 2011 01:19:04 am Wolfgang Denk wrote:
> Dear Aaron Williams,
> 
> please always keep the ML on cc:
> 
> In message <4D2FA7F0.7020900@caviumnetworks.com> you wrote:
> > It is based off of 2010.09 but I have applied all of the latest USB
> > updates from GIT with the exception of the multiple LUN support (this
> > breaks the SanDisk Cruzer because the Cruzer reports max lun=1). All
> > caches are enabled and this is on a 1.2GHz MIPS which is cache-coherent
> > (no need to flush/invalidate).
> 
> Maybe there are other issues with your code.  It's hard to comment on
> onknown out-of-tree ports.
> 
> Best regards,
> 
> Wolfgang Denk