Where do you get the 100k write cycles (isn't it erase-cycles, rather?) figure from? I've seen people use a lot of different figures, but it always seems to be pulled out of thin air... I do guess standard flash card memory might have somewhat fewer cycles than the superexpensive solid-state disks, though?

Interesting you say performance is faster than hard drive, read and write performance is usually a lot lower than modern drives, when you're talking standard flash cards... most CF/IDE adapters I've seen weren't even udma capable. Even one of my several year old drives (pretty slow by today's standards) can do 40mbyte/s sustained read, 25mbyte/s sustained write. My raptors are more like 85mb/s read, 60mb/s write. How fast is this adapter? (Of course also depends on the CF type used).

If you put linux on, you might want to consider using a logger daemon that sends log to a syslog server somewhere on the network, to reduce writes...

my CF is only 10MB/s but it is more reactive, the latency is very interresting.
when i said faster that hd, it was about the latency.
i don't have to wait and hear the noise. but it is very slow to read/write on it.

nothing is perfect!

100k writes is standard for Flash technology ICs. You can also get 1M writes with more expensive version. Cheapo USB Flash drives are mostly 100k cycle versions.

The higher priced cheapo models often use a wear levelling algo to increase the usage. The lower priced cheapo models often are just direct write. You need to check you model to see what it uses, but usually that info is difficult to find.

Ugh, models without wear leveling would be bad no matter how many cycles they support - at least if used with a traditional filesystem and workstation loads

I have read stories about Flash drives being rendered unusable in less than a day when used for DB storage on a server.

Ugh, sounds like they should have used better flash and a massive memcached - I do wonder how today's (expensive...) SSDs handle, though. And especially how the new flashram type that... intel+IBM, iirc?, are working on will improve things.

Btw, once you hit the max write cycles, can you still at least read the flash? That would still be a massive improvement over harddrives with headcrash.

The max. writes is not a hard limit. Usually at 100k everything should still be working fine, but after that you may find the occasional bit here and there that won't write/erase properly (it takes a long time to write/erase) until eventually the bit can no longer be programmed. More and more of these "dead" bits appear as the bit's write cycles increases. Reading is never a problem, just keep reading as much as you like.

a good design about system is the one used in our head.
the data never disappear of our brain. a cell cannot be overwriten at all.
only few neurons can change their state to access other banks of memory, but these one are RAM analogy, can be read/write infinitelly.
then, a good future system would use a Flash ROM for the Programs, assuming they are all mature, and don't need a lot of updates.
then, the datas will all be writen in fixed locations and never change, or rarelly.
musics never changes. for it, flash is the best solution as it doin't die in case of shock or hardcore read use.

then, it is interresting to have specific X86 machines with this kind of concept.

this one on the picture is my prototype for all OS construction. there are still hard drives for the Good old 98 , that is used as a transition.

and as times goes, it will become a full optimised computer for MP3 music and asm coding.

yeah!

http://en.wikipedia.org/wiki/Logfs
http://logfs.org/logfs/

BTW, I think Puppy Linux is now designed to (mostly) minimize writes to bootable USB drives, but I've never really tried installing to USB (and don't use it really, just mentioning it for completeness' sake).

Linux would write to it, for swop file use, you could try DexOS or MenuetOS, OctaOS, SolarOS.
DexOS would only write to it if you want it to.

interresting... if you have some dexos that support the creative SB PCI 128, i am really interrested to devellop for it.

what i waht is very simple. a fast MP3 player made in asm and "easy" to modify, using fasm and etc etc...

menuet is too much dirty for me...sorry ville, it would be with a lot of enjooyment that i join your os, but i cannot read it's source.

why not octa OS? because i've testedd it, and it freeze...
why not solarOS? because i've tested it, and it freeze too. but it is more in my philosophy, OO gui, very interresting...

instead to devellop my os, i will be a beta tester... and the winner will be... TG for his exelent tool!

There's a SB Pro driver if 128 as compatibility ?. as for asm mp3 code if you find any let me know .
Good luck.

Ps: i ended up buying one of these instead of coding a mp3 player
http://apple.clickandbuild.com/cnb/shop/ftdichip?op=catalogue-products-null&prodCategoryID=55&title=VMUSIC

i would prefer to code the MP3 codec, even if it take more than one year to do it.
after, it is ok, ones the code is done, it will work for the eternity of the target!

let me know if i can have the DEXOS source and SB driver for development on a HD.

i'll be happy to try to dev for MP3, as MP3 is one of the file format we cannot ignore.

Linux only does swap if you enable swap, and you don't want to go there for embedded use anyway...

Cool, i will upload and PM you a link to the latest source code.

Yes we can ignore,there is wma,ogg etc... And i like to listen to music with
a portable mp3 player ,not with the computer.

>i will be a beta tester
Not very helpfull if you don't report tests results.