<div class="gmail_quote">On Thu, Aug 19, 2010 at 9:21 PM, John Watlington <span dir="ltr"><<a href="mailto:wad@laptop.org">wad@laptop.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class="im"><br>
On Aug 19, 2010, at 1:59 PM, Tiago Marques wrote:<br>
<br>
> Thanks for the update. Can you disclose the brands of the mentioned SD cards?<br>
> It's always useful to know what not to buy.<br>
<br>
</div>I probably could, but variations between models from one manufacturer<br>
may be greater than variations between manufacturer. I suggest that<br>
anyone thinking of purchasing a number of SD cards test the ones they<br>
are considering buying, instead of relying on outdated test data.<br>
<div class="im"><br></div></blockquote><div><br></div><div>Ok, thanks. I'm half way down that path already, wanted to save some work.</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class="im">
> I'm assuming none of these cards don't have static wear levelling. Did any manufacturer provide you details on that or are they only using spare blocks for "repairability"?<br>
<br>
</div>They all have dynamic wear levelling, where blocks are actively moved<br>
in order to balance the wear across all block in the device.<br></blockquote><div><br></div><div>Nice to know. Unfortunately I've come across some flash devices that seem to have none.</div><div>Correct me if I'm wrong. Dynamic wear levelling is the one done on the free blocks and not the ones that have data, right? </div>
<div><br></div><div>From your tests, 10TB on 2GB cards would imply 5000 write/erase cycles with static wear levelling, which is pretty good. An SSD I bought a while ago also had 5000 W/E cycles and that number seems to be going down as they migrate to smaller process nodes. Perhaps that's the case?</div>
<div>3 bits per cell MLC also doesn't spell good things IMHO, especially as they're only reducing the die size in 20% for the same capacity(as per Intel/Micron 25nm 3bpc flash).</div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<div class="im"><br>
> Any plans for the use of NILFS? It seems to help with wear levelling, although<br>
> it is not a completely circular log FS.<br>
<br>
</div>We are starting to look at it.<br>
Does anybody have experience shipping it ?<br>
even using it ?<br></blockquote><div><br></div><div>I've stuck with EXT2 with noatime for now but was looking into it. The only thing I didn't like was it being not fully circular and being arranged in blocks. Other than that it seems fine and should help a lot with the random write latencies, as you may know, which was my motivation for looking into it.</div>
<div><br></div><div>Best regards,</div><div>Tiago </div><div> </div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
<br>
Cheers,<br>
wad<br>
<br>
</blockquote></div><br>