YES I ever measured and improved the sharing performance during
development.
What you has pointed out match our current design:

1. There is concurrent thread, if UI activity is launched as visible.
One is writing to database about the sharing progress, and the other
querying to update the UI. So during sharing, performance is better if
just keep OPP session in background.
Even if user doesn't launch UI, there is still frequent DB updating,
which is time consuming too.
I'm sure there could be improvements for the concurrent cases in app,
but we can't just delete the DB operations.

2. I believe the 16k you mentioned is the read buffer size for file in
BluetoothOppObexClientSession.java
BufferedInputStream a = new BufferedInputStream(fileInfo.mInputStream,
0x4000);

This 16k buffer size is a balance of local buffer size and remote
buffer size, although 16k might not be a perfect number.
Remote may use different OBEX MTU, like 4k, 8k, 16, 32k, 64k, etc. New
devices(mostly smart phones) may use 64K because of its high
performance, while most legacy phones may use 4k.

I think your data is collected when use a device with 64k OBEX MTU.
In this case, we have to transfer 4 times to fill remote buffer, which
match your result.
But if we are transferring to a 4k buffer device, larger buffer size
will be split into more packets, and actually slow down the transfer.

Thanks for your data. Let's improve it continuously.

On Mar 5, 4:01 am, Bheemsen <[email protected]> wrote:
> Hi All,
>
> I am using Eclair 2.1 on OMAP platform.
> Eclair 2.1 has BT-OPP support (file transfer from Gallery application)
>
> With OPP application throughput numbers around 600 Kb/s (tested with 15M
> file).
>
> We found that the UI updation is taking around 40-50ms for every 16k data
> transfer.
>
> 1) If i comment the UI updation the throughput comes around 950kbps.
>
> 2) In BluetoothOppObexServer.java the read size is 16 Kb/s ,if i increase
> the read size from 16Kb to 64 Kb the throughput will be 820 Kb/s.
>
> Is anybody measured the throughput with OPP?
>
> Is there any way to improve the throughput?
>
> --
> Regards
> Bheemsen

-- 
unsubscribe: [email protected]
website: http://groups.google.com/group/android-porting

Reply via email to