Feeding SSDs from +5VSB for outage data corruption safety

[TELVM]

Re: Feeding SSDs from +5VSB for outage data corruption safety

ok.... so where will the command to flush the cash to the disk come from?

Aaaargh you might have shot me down in flames .

... During a ‘clean' shutdown, a host system will initiate the ATA STANDBY IMMEDIATE command to the SSD and give it enough time to prepare for the shutdown. This substantially enhances the ability of the SSD to save data which is currently in transition or in-flight (in temporary buffers) to the non-volatile NAND Flash media.

During an unsafe power shutdown or a loss of power in SSDs that do not have this PLI feature, the SSD abruptly loses power prior to receiving the ATA STANDBY IMMEDIATE command. Without this critical command or other PLI protection mechanism, the SSD would lose any data which is sitting in temporary buffers when the power is lost ...

http://www.intel.co.za/content/dam/w...logy-brief.pdf

If system goes belly up under an outage without issuing STANDBY INMEDIATE, the consumer SSD maybe dumb enough to continue business as usual during the miliseconds of grace provided by the caps, instead of flushing ASAP, equally losing data in the end .

[tom66]

Re: Feeding SSDs from +5VSB for outage data corruption safety

Solution, although difficult to implement, is a circuit which monitors incoming mains and alerts the kernel to this event.

[TELVM]

Re: Feeding SSDs from +5VSB for outage data corruption safety

... An “unexpected power loss” is an event where a power loss occurs and is not preceded by an ATA STANDBY IMMEDIATE or similar command. STANDBY IMMEDIATE is a system-level command which alerts the storage device to an impending loss of power, or entry into low power modes like SLEEP and HIBERNATE.

STANDBY IMMEDIATE is frequently not sent in a power loss event such as an unexpected loss of power connection, an expired battery, or holding down the power button for 4 seconds or longer ...

http://edge.crucial.com/firmware/m4/...ility_040H.pdf

Guess this invalidates the idea .

[goontron]

Re: Feeding SSDs from +5VSB for outage data corruption safety

Aaaargh you might have shot me down in flames .



If system goes belly up under an outage without issuing STANDBY INMEDIATE, the consumer SSD maybe dumb enough to continue business as usual during the miliseconds of grace provided by the caps, instead of flushing ASAP, equally losing data in the end .

Guess this invalidates the idea .

sorry

[tmiha71]

Re: Feeding SSDs from +5VSB for outage data corruption safety

Idea has sense, heavily counts on disk behavior - when no reading is performed (or asked), it's time to do writing (if there's any pending)...

But question is if that is real disk behavior...

As stated above, no communication with host - what will hard disk do ?

[goontron]

Re: Feeding SSDs from +5VSB for outage data corruption safety

Idea has sense, heavily counts on disk behavior - when no reading is performed (or asked), it's time to do writing (if there's any pending)...

But question is if that is real disk behavior...

As stated above, no communication with host - what will hard disk do ?

i can tell you what most seagate barracuda HDs will do, with certain variables of course!

if APM is enabled and if there is data in the cache, once the timeout is reached it will write data to the disk and after what is in the list of files to be read is read, then that data sits in cache and the drive spin down, but thats IF the caching mode flag is there. else if there is no cache flag, as the commands come in, they get queued, then run, then the timeout counter starts.
but you cant disable cache on a SSD, the controller will tell the computer it is disabled, but its not. you read how a SSD writes things for wear leveling? how, it remaps "sectors" and call's them other sectors? that's why it wont disable cache!

[TELVM]

Re: Feeding SSDs from +5VSB for outage data corruption safety

... As stated above, no communication with host - what will hard disk do ?

Enterprise SSDs with integrated protection detect the voltage drop in the rail during an outage, and self-administer the STANDBY INMEDIATE command.

With client SSDs without integrated protection, I honestly don't know what exactly would happen if communication with host is lost, but power keeps on for some milisecs courtesy of some DIY concoction. But in principle I'd expect them to just run on as usual without doing anything special, and lost data in the end equally .


Integrated outage protection is beginning to reach client SSDs, for instance the cheap Crucial M500 has a troop of SMD caps to sustain emergency cache flushing:



We can also make lemonade and use some SSD with Sandforce SF-1000 series controllers, which don't use DRAM cache, hence have nothing to flush, and are thus practically impervious to unexpected power losses.