I am guessing the write cycle numbers are pr bit, as unless the USB stick has crap all wear leveling it can survive way more than 3000 cycles.
Frankly the only documented case i recall is of someone killing a USB stick was back during the early MB days, when someone mounted it with the sync option under Linux. This, in combo with it being FAT formated, resulted in the bits holding the allocation table getting a whole lot of writes.
Without the sync option, the stick would probably have survived for quite some time as the table would get updated less frequently.
These days you are unlikely to see such a problem unless you are scraping the bottom of Ebay listings.
It depends on what you're writing and where you're putting those writes. Flash drives survive this level of activity because they intelligently queue up the writes and don't aggressively re-write cells unless it's necessary.
The USB stick reference is mostly bad tech journalism. For raw flash media itself, 3000 cycles is in the right ballpark for multi-level NAND Flash. For whole USB sticks or SSDs, many mechanisms are in place so that a user doesn't need to worry about the specific 3000 cycle life of a specific chunk of flash.
Originally 2006, then 2008, then 2011, 2012, 2014, and 2015... It was officially canned (and by that I mean moved from memristor based to traditional architecture) last year.
RIP HP Labs being an actually interesting place for computer architecture research.
"HP has abruptly changed course on its 'Machine,' a new type of memory-driven computer it thinks will radically alter large-scale data processing. When the company first launched it last year, the plan was to use a new kind of memory chip called the "memristor," which is as fast as DRAM but can permanently store data. The problem is that the tech, which HP expected to commercialize with Hynix in 2013, still isn't ready. Rather than giving up, though, HP has decided to take it in another direction by using both conventional RAM and phase change memory."[1]
To translate the article, "the tech [..] still isn't ready" means management thought it would be cool to build memristors but they have no clue how. Some engineers will be fired. They came up with a new idea, the Machine instead. They will probably get a bonus.
Surprisingly, promises to invent never seem to materialize. See thegrid.io.
I'm reminded of the old saying: The stock market has forecast nine of the last five recessions. New memory tech always seems to be just around the corner, if I had a dollar for every time it was predicted I could retire.
But by then, the feature-phone business was in fast decline, and I guess there wasn't enough demand for continued investment that technology vs. more investment in higher capacity NAND for the smartphone market.
imagine if ssds didn't actually exist as a consumer product but was just constantly brought up by large companies as the "next big thing!" that never actually came out.
i mean intel, and hp all have their own versions and they've been working on them for years and we've yet to see anything come of it. i'm not holding my breath for this to actually get released.
I've seen a lot of 'looks promising... can't make work or can't make economic' technologies over the years. Just the nature of the business. I'd hate for companies to stop trying tho. Technologies that seek to displace another older technology often have a long road ahead. Because no green field, the economics and performance has to beat the established technology before anything sees anything but niche applications.
I think it took ten years for NAND flash to start seriously chewing into the mass storage market
28 comments
[ 5.0 ms ] story [ 68.7 ms ] threadFrankly the only documented case i recall is of someone killing a USB stick was back during the early MB days, when someone mounted it with the sync option under Linux. This, in combo with it being FAT formated, resulted in the bits holding the allocation table getting a whole lot of writes.
Without the sync option, the stick would probably have survived for quite some time as the table would get updated less frequently.
These days you are unlikely to see such a problem unless you are scraping the bottom of Ebay listings.
RIP HP Labs being an actually interesting place for computer architecture research.
[1]http://www.engadget.com/2015/06/05/hp-the-machine-no-memrist...
Surprisingly, promises to invent never seem to materialize. See thegrid.io.
But by then, the feature-phone business was in fast decline, and I guess there wasn't enough demand for continued investment that technology vs. more investment in higher capacity NAND for the smartphone market.
i mean intel, and hp all have their own versions and they've been working on them for years and we've yet to see anything come of it. i'm not holding my breath for this to actually get released.
I think it took ten years for NAND flash to start seriously chewing into the mass storage market
This research seems to take program/erase cycles into greater account.
Sadly today servers take a long time to boot (up to 6 minutes) making this ineffective for a baremetal server in that must give five nines uptimes.