CompactFlash write speed

Hi all!
I'm designing a high speed datalogger that should store
a continuous data stream (about 2MB/s) for about 30
minutes. I'm using a Compact Flash card to store data
samples (no FAT, just raw data with a continuous
addressing, in order to keep all as fast as possible).
The CF I tested is rated 66X (i.e. about 10MB/s),
and the micro (an AVR) can sustain a write speed
greater than 4MB/s. The problem is that the CF
seems to have very long execution time, so even
if I'm very fast to transfer commands and data
blocks, I have to wait *many ms* before getting
the card ready again for the next command.
My question is: does anybody succeded in getting
a sustained write speed greater than 2MB/s with
a CF card? If yes, how? What is the most appropriate
sequence of commands/operations: write sector,
write sectors, erasing before writing, writing without
erasing, using contiguous addressing, or what else?

Thanks in advance,
Antonio



P.S. I'm using True IDE mode, 8 bit data, PIO
mode (not DMA). But as I said the limit is not
the interface but the card "execution" time...

A.D. wrote:
> Hi all!
> I'm designing a high speed datalogger that should store
> a continuous data stream (about 2MB/s) for about 30
> minutes. I'm using a Compact Flash card to store data
> samples (no FAT, just raw data with a continuous
> addressing, in order to keep all as fast as possible).
> The CF I tested is rated 66X (i.e. about 10MB/s),
> and the micro (an AVR) can sustain a write speed
> greater than 4MB/s. The problem is that the CF
> seems to have very long execution time, so even
> if I'm very fast to transfer commands and data
> blocks, I have to wait *many ms* before getting
> the card ready again for the next command.
> My question is: does anybody succeded in getting
> a sustained write speed greater than 2MB/s with
> a CF card? If yes, how? What is the most appropriate
> sequence of commands/operations: write sector,
> write sectors, erasing before writing, writing without
> erasing, using contiguous addressing, or what else?
> 
> Thanks in advance,
> Antonio
> P.S. I'm using True IDE mode, 8 bit data, PIO
> mode (not DMA). But as I said the limit is not
> the interface but the card "execution" time...

I am using the compact flash with BlackFin system and with the good 
cards I am getting the sustained read/write performance at the order of 
25 MBytes/sec.  Just about every modern compact flash card from WalMart 
can do 10MB/s at least. I am not using the DMA/UDMA modes, all done with 
the conventional sectors read/write operation.  The 25MB/s is the 
limitation of the system; the CF itself can go faster then that. Yes, 
there is a long latency at the beginning of the commands. For the best 
performance, you should use the maximum block size, and align your 
read/write flash operations on the block size.


Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com

Vladimir Vassilevsky <antispam_bogus@hotmail.com> wrote in message
X95gk.14673$xZ.5944@nlpi070.nbdc.sbc.com...
>Yes,  there is a long latency at the beginning of the commands.
> For the best performance, you should use the maximum
> block size, and align your read/write flash operations
> on the block size.

Thank you for the info!
Surely I'm doing something wrong, since I observe
huge latencies (up to 200ms!)... Maybe these are
something like timeouts and not command latencies...

Thank again,
Antonio

>Vladimir Vassilevsky <antispam_bogus@hotmail.com> wrote in message
>X95gk.14673$xZ.5944@nlpi070.nbdc.sbc.com...
>>Yes,  there is a long latency at the beginning of the commands.
>> For the best performance, you should use the maximum
>> block size, and align your read/write flash operations
>> on the block size.
>
>Thank you for the info!
>Surely I'm doing something wrong, since I observe
>huge latencies (up to 200ms!)... Maybe these are
>something like timeouts and not command latencies...
>
>Thank again,
>Antonio
>
>
>
Hi Antonio,
Is this problem resolved?
I am also having the same problem. Most of the time the write takes
200-300 ms. I am running this in true-IDE mode and using DMA for the data
transfer. DMA happens very quickly but the CF will be in busy state for
more than 200 ms.
Any help really appreciated

Thanks
Prasanna


>
>
>

prasannahm <prasannahm@hotmail.com> wrote in message
ScOdnZUi_-n-ak3VnZ2dnUVZ_rjinZ2d@giganews.com...
> Is this problem resolved?
> I am also having the same problem. Most of the
> time the write takes 200-300 ms. I am running
> this in true-IDE mode and using DMA for the
> data transfer. DMA happens very quickly but
> the CF will be in busy state for more than 200 ms.
> Any help really appreciated

Hi Prasanna,
all CF I tested seem to have a huge busy period
(200/300ms) following the very first writes, but
then it will decrease somewhat to an avarage
of few ms. These busy periods and their length are
quite unpredictable (even using the same commands
in the same sectors), but are strongly dependent
on the specific card brand and speed rating.
When using faster and more expensive cards
(>133x) busy periods are shorter are less frequent.
However, in order to avoid the problem the only
solution is to use lot of buffering...

Bye,
Antonio