Re: USB high speed
- From: "Wei Chen" <weichen@xxxxxxxxxxxxx>
- Date: Mon, 11 Dec 2006 09:14:05 -0800
Hi Jim:
Could you please tell us more about how you conducted your performance test? Is it done by calling ReadFile/WriteFile at filesys level, or read/write sectors at block driver level, or just copy/paste files through UI?
The performance of throughput based on USB2.0 is heavily depending on the data payload size --, What we see internally is that, in order to get a throughput around 50Mbps, you probably need to use the payload size around 8M Bytes. There are some significant improvements in CE6.0 though.
If you measured the performance by copy/paste files through UI, then there is a limitation on the payload size enforced by filesys (when filesys issues read/write operations down to storage driver level). In such case what you see (~2MBps) is probably near the limit.
Thanks,
-Wei
"JimAA" <JimAA@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message news:D23F081C-C470-4A2D-93C1-11CEBD8AFBAD@xxxxxxxxxxxxxxxx
We're using USB Host (PC) with a USB hard-disk as the client. Both directions
are slow. We've been able to get about 30 Mbps max by installing CE 5.0 and
then some of the patches. We're using the entire CE USB stack (SBP-2, etc) to
read/write files to/from the drive. I wonder if there's an example somewhere
that can be used for benchmarking.
"Steve Schrock [MS]" wrote:
Are you referring to USB Host or USB Function? Which client are you using?
Which direction appears to be slow--IN or OUT or both?
I have seen USB 2.0 throughput well over 120 Mbps from CEPC to CEPC for both
BULK IN and BULK OUT using a simple test function client on one side and a
test host client on the other. Of course once you start using a meaningful
protocol instead of the test client, that number will decrease.
--
Steve Schrock
Windows Devices Core BSPs
This posting is provided "AS IS" with no warranties, and confers no rights.
"JimAA" <JimAA@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
news:AE65B442-1437-452F-91A1-581CA6B407E8@xxxxxxxxxxxxxxxx
> That's a good point. It could be a driver problem. We're using a BSP
> provided
> by Freescale for their iMX31. However, we did test WinCE running on a > PC
> and
> got the same slow results.
>
> It would be nice to find another driver to verify against.
>
> "Steve Schrock [MS]" wrote:
>
>> What scenario are you testing? What data leads you to believe that >> this
>> is
>> an OS problem rather than a driver problem in the BSP?
>>
>> -- >> Steve Schrock
>> Windows Devices Core BSPs
>>
>> This posting is provided "AS IS" with no warranties, and confers no
>> rights.
>>
>> "JimAA" <JimAA@xxxxxxxxxxxxxxxxxxxxxxxxx> wrote in message
>> news:FE3B1B01-3565-4D1F-B696-7ABB1F32CD5B@xxxxxxxxxxxxxxxx
>> > We're evaluating Windows CE of use on an ARM processor and finding >> > that
>> > the
>> > USB 2.0 (high speed) is very slow (about 2MBytes/s). The same >> > processor
>> > running Linux is 23 MBytes/s.
>> >
>> > Has anyone experienced this kind of poor performance from CE and is
>> > there
>> > a
>> > fix?
>>
>>
>>
.
- References:
- Re: USB high speed
- From: Steve Schrock [MS]
- Re: USB high speed
- From: Steve Schrock [MS]
- Re: USB high speed
- From: JimAA
- Re: USB high speed
- Prev by Date: Re: Where is cmd.exe on my Windows Mobile 5.1?
- Next by Date: Re: Telnet to Modem/Bridge
- Previous by thread: Re: USB high speed
- Next by thread: Re: WM5 Chinese characters
- Index(es):
Relevant Pages
|
