Мисля че флаш паметта е в GB не в MB.

[Отговори на този коментар]

Не знам дали хората са мързеливи или не знаят.

http://en.wikipedia.org/wiki/Sinomanic

Цена средно около 250$ за което получавате почти МИПС съвместимо китайско чудо и модифициран дебиан.



[Отговори на този коментар]

Към: Я бе!
От: hydra <mhydra __@__ users< dot >sourceforge< dot >net>
На: 17-08-2007@6:40 GMT+2
Оценка: 1/Неутрален

Браво на китайците. Те отдавна казаха че правят техен собствен процесор но не си мислех че ще го пуснат в продажба толкова скоро.


[Отговори на този коментар]

---

Какво ще кажете за устойчивостта на флаш паметта — би ли издържала голям брой презаписи? Как е в сравнение с харддисковете — по-устойчива или по-неустойчива в това отношение?

[Отговори на този коментар]

Към: Какво ще кажете за флаш паметта?
От: the_real_maniac
На: 17-08-2007@8:05 GMT+2
Оценка: 1/Неутрален

Много зависи от типа на Флаш паметта, има такава с

10г и 100.000 записа
има такава с 40г и 1.000.000
Гледам и 10.000.000 презаписа някъде,но като цяло според мен в това отношение просто са назад , защото ако пишеш/променяш само на едно място и ... познай :-) не bad sectors бла-бла , ами ще имаме нов термин old bits/mem blocks:-D (а един файл като се създаде си стои на определен адрес/място в по-голямат си част дори и да се променя)

Но и аз все се каня да видя точно кой тип ползват
сигурно е NaND flash, но като гледам повечето са 100.000 +/- ;-) (сайта на intel , на micron може да видиш ,толкова са nand-четата, но логично ползват следната хитрина, уравновесяват кой блок по колко пъти е бил записван с цел (уж) увеличаване на животa, но общото на цялото като едно си е 100.000 , т. е на цялата flash памет(ако е 1 GB) наведнъж от до .... по толкова 100.000 пъти).

ето още инфо:

intel nand - http://download.intel.com/design/flash/...

micron nand - http://www.hynix.com/eng/02_products/03...

цитат от http://forums.ds-xtreme.com/archive/ind...


mmmm......future proof........mmmmm......lemme think.........flash memory.............100,000 r/w cycle.......... 1+1=
(correct me on the r/w cycle)

Ok, I will.

First, "r/w cycle" is not the correct term. It is more correctly "P/E cycles" or Program/Erase cycles. Reads do not affect the wear of the flash, only erasing and reprogramming does.

Second, it should be pointed out that this cycle count is per "block." NAND flash is made up of a collection of individual blocks, each block of which is, in essence, a small individual flash memory. We are using what is referred to as "SLC Large Block" which basically a fancy way of saying that each block is 128KB, and each page (the smallest unit of a P/E cycle) is 2KB.. This means that there are 4,096 blocks in the DS-Xtreme flash.

Each block has a 100,000 P/E cycle count, not the flash as a whole. In our hardware logic, we perform a nice little trick called "dynamic load leveling." What this does, is ensures that each block in the nand is written to in sequence, and re-mapped within the logical address space. For example... let's say that we have data sitting at "physical block" 136. We want to rewrite that data... so we erase block 136, but instead of writing the new data back to block 136, we check our load-leveling table, and see that the next block "in the queue" is 5,736... so we write the new data there instead. We then update our load leveling table with the next free block for our next write... our original block 136 will not be written to again until EVERY other free block in the flash has been written to once. This means that we can issue 409,600,000 writes before a single block will fail.

Just to take things a step further.... NAND flash, unlike classic NOR flash, has a funny little quirk. When the flash comes out of the factory... some blocks will almost always be bad right from the start. So the flash manufacturers came up with a neat little trick.... They simply stick extra blocks in. They also know that, down the road, more blocks can fail... so at the end of each block, they include a little "Error Correcting Codes area" which allows us to know when a block is failing. So... even if you do reach that 410 millionth write, and a block fails... that block is simply marked as bad, never used again, and replaced by one of the spare blocks. There are (if i recall correctly offhand) 128 of these "spares" adding (approximatly) another 1,280,000 write cycles.

It should also be mentioned that this is not a deterministic number, it is simply the "guaranteed minimum" from the flash vendor. It does not mean that on the 100,001th write to a given block, that block will simply die. There is nothing to say that a block is even assured to die at some point, regardless. It is quite feasible that any given block may simply never die. This is simply the number they are willing to put down on paper as their "you can sue us if your parts fail before this point" number. As I'm sure you can guess... they've probably low balled it *a lot.*


So how long does this really give us? 410,880,000 writes of 128KB each... at a throughput of .2 seconds per write (a rough timing i just took from my mac) it would take a MINIMUM of approximately 951 days (almost 3 years) of CONTINUOUS round-the-clock bombardment of program/erase cycles before things started failing. My guess? It would probably take more like 8-10 years. Why don't you try it out, and come let us know when it starts failing? :D

[Отговори на този коментар]