Date: Tue, 8 Oct 2013 02:15:36 +0400 From: Anatoly <anatoly@kazanfieldhockey.ru> To: Julian Elischer <julian@freebsd.org> Cc: freebsd-usb@freebsd.org Subject: Re: hot usb sticks Message-ID: <20131008021536.43345a8e@asd.localdomain> In-Reply-To: <8325A99D-54C4-4EE0-968B-6605665F930F@bsdimp.com> References: <201310051335.r95DZOx4004869@fire.js.berklix.net> <5250344E.2000500@quip.cz> <5252B9E4.2050208@freebsd.org> <8325A99D-54C4-4EE0-968B-6605665F930F@bsdimp.com>
next in thread | previous in thread | raw e-mail | index | archive | help
В Mon, 7 Oct 2013 08:07:48 -0600 Warner Losh <imp@bsdimp.com> пишет: > > On Oct 7, 2013, at 7:40 AM, Julian Elischer wrote: > > > On 10/5/13 11:46 PM, Miroslav Lachman wrote: > >> Julian H. Stacey wrote: > >>> Has anyone else noticed how hot USB sticks can get when used for > >>> backup ? & also that IO errors occur after a while, which go > >>> away after a cold reboot. > >>> > >>> Not the whole stick, but the metal connector gets hot, so chip is > >>> hotter still. Obviously one won't notice this on large plastic > >>> encassed sticks, but 2 main sicks I use are: > >>> sandisk 2Gig metal case "vendor" "0x0781"; "product" "0x5151"; > >>> delock 8G miniature (~ 3mm of platic beyond plug) > >>> "vendor" "0x05e3" "product" "0x0727" > >>> > >>> I usually notice this when I am updating (writing) a crypted > >>> (gbde) UFS file systems using port/net/rdist6 (which only > >>> rewrites updated files). > >>> > >>> Source data is 1,446,438 K bytes in 42,611 files so average > >>> size of 34 K. But a lot of the files are really small, (~/.* > >>> config & mail files etc, so as rdist will be updating each one > >>> sequentially, & each will take a read + write cycle on a stick > >>> block,& as many small files will probably map to the same stick > >>> block, thats some concentrated cycles. > >>> > >>> More stick detail at > >>> http://www.berklix.com/~jhs/src/bsd/fixes/FreeBSD/src/jhs/etc/devd/jhs.conf > >>> > >>> Quite often I have to reboot my target host that has a stick > >>> inserted, I believe regardless of OS version on USB target host > >>> > >>> Possibly there might be less heating when only reading (as read > >>> cycles are also quicker), but mainly I'm backing up, writing. > >>> > >>> I was thinking of making a heatsink to clamp to a USB socket on an > >>> extension cable, but before that I'll try hanging a USB extension > >>> cable adjacent to a case fan. > >> > >> I have a few USB sticks, some of them are really old (and fast!), > >> for example 512MB A-Data with 200x speed, or 8GB 133x. These fast > >> sticks are almost cool. Some cheap modern sticks are hot even if > >> used as read-only for booting ZFS backup server, where whole base > >> system is on UFS USB stick monted read-only and all writes are on > >> ZFS partitions of 4 HDDs. Even in this RO scenario, the hot stick > >> died after about 2 years. Writes on it was made about 3 times a > >> year because of system or ports updates. > >> > >> So in my case: newer -> cheaper -> slower -> hotter = shorter life. > > > > actually, hotter is not always worse in Flash. > > Warner can say more in detail but hot is good in some case while > > cold is best when you put the device on the shelf. > > The amount of damage that a P/E cycle does to a cell often is > determined by the dwell time and temperature between program and > erase. There's a well known Arrhenius relationship here, so as the > temperature gets hotter, the effective dwell time increases, which > allows more of the damage to heal. So heat can be good in a high > write scenario. > > However, in a low write, long retention scenario, high heat, by the > same Arrhenius factor, will lead to much shorter retention, which may > have contributed to your failure. There are certain housekeeping > activities that go on in the USB stick, typically, that will move > really old data, which causes additional writes to the flash that you > are unaware of. Also a high read work load can tax the cells, which > have only so many reads per page specified by the manufacturer. > > But if the errors occur while it is HOT and not while it is COLD, I'd > be suspecting an issue of construction that's to blame (eg, bad > design that works at low temps, but at high temps the electrical > characteristics of the data transmission lines change just enough to > be out of spec). If it just died, then the high temperature > accelerated the aging effects beyond the bounds of the drive firmware > to cope. > > Warner I've seen such a way faulty pen drives (IO errors after sustained write, or even short-term writes, but repeated many times for a long period), however can't say they was hottest I've seen. But they was almost cheapest. Also I've noted, say 8G sticks made at days when they was priced 30$ is more reliable than those made at days they priced 6$ Anyway I using my "choosen thrusty" sticks and SD cards for backups and running OSes with no problem. I just want to remind what we can do to minimize wear (and speed up): -align filesystem with flash memory erase block. 8192 sectors is almost good fo any flash media. Note such offset on new factory-formatted sticks. Before I get it, I was wonder why re-partitioned pendrive becomes slower than factory-partitioned. read: http://forums.freebsd.org/showthread.php?t=19093 -mount with noatime option if updating file access times is not necessary If I need to backup large amount of small files, for me, sometimes it's faster to create "image" of pendrive somewhere on local HDDs, mount it, copy files on it, and then dd it to pendrive.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20131008021536.43345a8e>