General > General Technical Chat
Dollar Store USB Sticks
<< < (3/3)
msuffidy:
I think if you scan instead of just clear and a few other options is basically works. I tried to change a few options but I get the feeling I did not do a relocate sort of scan. I am going to see if it changed anything. Oops I put a capital A in there. Oh well. I have a windows 7 vm in linux. I copied the .qcow or whatever to a new one and made a new machine with the copy. Then I put in that stuff that was sort of dubious and disabled the network so it could not do anything to my actual computer. When I run the VM I split up a lot of the hardware to the VM instead of the main computer, so the ports were directly interfaced to Windows 7. Well just got a diff error on 6GB of writes.
msuffidy:
I came up with an approach to make it worth something to me. I used the factory scan that I think differs from the stand scan in that it does not de allocate block. The theory is like this.

1) I see there is this ECC level, which was 12 of 15. I take that as an attempt to get good data out of the bits of the flash regardless if they are all working or not.
2) I would say sporadic failures occur with certain data in certain offsets in the memory that was not in its initial bad sectors list as per using ECC.
3) I argue if I can set ECC level zero that would eliminate any correction, and do a disk scan with patterns I can find the bas sectors.
4) Through software, granted certain critical areas are ok like the MBR block, I can make an exclusion map using ext4 for example.

I just did one test and my initial 3+Gb write diffed ok. So if this is the case I would say that the basic technology was reliable, but it was implemented in a non reliable manner. It could be the technology is fundamentally unreliable. NOPE second try produced lots of errors.


****
ok a few hours later. Well I noticed a few things. One is it will accept pretty much anything I want it to do if I clear the stick and then re install it before doing it. The last setting I wanted to see if I altered 'r/w' cycle to 66ns, that is to say try to slow down a process. I left ecc zero. What happened and it was rather odd was the first 3 Gugs again verified (without any bad block declaration). So I am still testing things.

****

Mar 7 - Still have not found a reliable configuration. It works for a while and then totally messes up, so I guess that is the way this stuff just is.
niconiconi:
Cheap USB flash drives are where all the bad NAND flash chips of the world went to...
msuffidy:
Well I think the point here is I TRIED to get a dollar store USB stick thinking it would be slow but may be data safe. That was not the case. I did a last ditch reflow in my literal oven and it started to melt the solder, but it was never really flowy. I was hoping to fix up bga connections. The reflow (attempt) did not make it reliable, but it still works. I thought I would make a final set of conclusions.

1) The idea that I could turn off ECC and do a bad blocks check through software may have been flawed by the fact reading may be compromised by error and thus I did turn the ECC back on.
2) My experiments showed that choosing the fastest access time of 25ns with my unit still worked (as well as any slower setting), thus longer accesses are just wasting time.
3) My final settings were  25ns r/w cycle, factory scan, ecc level 12.
4) I am not even sure the flash has problems because areas that before failed did succeed in getting written to later. The problem may be in the controller.
5) What seems to work is to insert the stick and do an filesystem image write to it immediately using dd. I have done a 12GB write and a 4GB write just now doing this (without diff errors). I noticed a pattern of the first writes mostly working as I was experimenting. So for that application at least I can use that method to give movies etc to friends to copy and thus be a use of some sort.
Navigation
Message Index
Previous page
There was an error while thanking
Thanking...

Go to full version
Powered by SMFPacks Advanced Attachments Uploader Mod