Personal computing discussed
Moderators: renee, morphine, Steel
Waco wrote:Production in what sense? What kind of availability do you need? What's the average GB writes/day?
Without those it's pretty hard to gauge whether you could get by with consumer products.
UberGerbil wrote:I can't say for sure without looking at the specs for whichever enterprise drive you're considering, but I wouldn't assume "factory" overprovisioning is the only difference.
...
NAND binning, ECC, controller logic to further reduce write amplification, etc) some of which may also enhance endurance (and others -- like encryption or backup power -- which may not)
UberGerbil wrote:They all have ECC, but not all ECC is created equal -- if you have more bits to devote to it and/or more processing power in the controller, you can cope with more entropy. I'm on my way out the door, and don't have a lot details to pass along anyway, but StorageReview has been doing Enterprise SDD reviews for a while and may be able to find some of the details you're looking for either in their reviews of particular models or in their forums.
Ringofett wrote:Regarding NAND binning and longevity, I'd point to TR's very own endurance review. I've no doubt commercial SSD binned NAND might last a little better, but holy cow, they've already pounded those with mind-boggling, warranty-voiding numbers of writes and only a handful of bad sectors, not a single bad die, and not a single drive failure. So, I don't know that it'd matter, unless you're at the LHC or someone similar producing petabytes and intentionally will abuse your drive for years.
As for other features, I'll cede to others, except to say as far as ECC goes, I love btrfs/zfs/ReFS as a second line of defense for critical files vs just built in ECC.
Captain Ned wrote:Are you thinking that if you leave 100GB outside of the partition table on a consumer drive the firmware will add that unpartitioned space to its pool of spare cells and thus "add reliability"? That'd take a custom firmware.
just brew it! wrote:@Cap'n - I think you're mistaken. If you set aside a partition, and (here's the crucial part) never write to it, then the flash controller should treat the cells which are initially assigned to that partition as part of the free pool. They will get used during wear leveling, just like any other other cells that have never been written (or cells that have been TRIMmed). The bottom line with underprovisioning is to never touch some percentage of the drive's capacity; the easiest way to ensure that is to create a partition you never use.
@DTD - Enterprise storage is typically tested more rigorously and has a longer warranty, and has firmware that is tuned for use in RAID arrays instead of as a single drive. Enterprise hardware may also use components with a wider operating temperature range. So these are all factors that may need to be considered as well, depending on your use case.
is there any way to reclaim this? A secure erase or similar?The bottom line with underprovisioning is to never touch some percentage of the drive's capacity
Duct Tape Dude wrote:just brew it! wrote:@Cap'n - I think you're mistaken. If you set aside a partition, and (here's the crucial part) never write to it, then the flash controller should treat the cells which are initially assigned to that partition as part of the free pool. They will get used during wear leveling, just like any other other cells that have never been written (or cells that have been TRIMmed). The bottom line with underprovisioning is to never touch some percentage of the drive's capacity; the easiest way to ensure that is to create a partition you never use.
@DTD - Enterprise storage is typically tested more rigorously and has a longer warranty, and has firmware that is tuned for use in RAID arrays instead of as a single drive. Enterprise hardware may also use components with a wider operating temperature range. So these are all factors that may need to be considered as well, depending on your use case.
Thanks, just brew it!. That's some very informative stuff. I thought the term was "overprovisioning" but if it's actually "underprovisioning," I hope you all know what I intended. I'll keep using "overprovisioning" for thread consistency though.
The RAID firmware is a good point, though I think for for my purposes I'd only be considering RAID 1--which I would imagine benefits less from this firmware than a RAID 0/5/etc. I could be wrong!
The operating temperature is a great point, I saw some press releases advertising better SSD endurance at 50C+. Do we know how hot the TR endurance benchmark is running? I can't find temps on the 300TB or 500TB updates. Judging by the setup here: http://techreport.com/review/24841/intr ... periment/4 I would think these are running pretty cool. Anyone care to guess? <30C? <40C?
Out of curiosity, when you say:is there any way to reclaim this? A secure erase or similar?The bottom line with underprovisioning is to never touch some percentage of the drive's capacity
Again, thank you all for your answers and discussion so far. I'm learning a ton!
just brew it! wrote:@Cap'n -
I think you're mistaken. If you set aside a partition, and (here's the crucial part) never write to it, then the flash controller should treat the cells which are initially assigned to that partition as part of the free pool. They will get used during wear leveling, just like any other other cells that have never been written (or cells that have been TRIMmed). That's how SSD wear leveling is supposed to work. The bottom line with manual underprovisioning is to never touch some percentage of the drive's capacity; the easiest way to ensure that is to create a partition you never use.
Captain Ned wrote:Consider me educated. For some reason I simply assumed that the firmware would only look to the spare cells built into the firmware at the factory.
just brew it! wrote:(I've seen a number of Linux users recommend disabling automatic TRIM in the OS, and running a daily or weekly batch trim instead. Not sure if there's an equivalent to this in Windows.)
Ryu Connor wrote:just brew it! wrote:(I've seen a number of Linux users recommend disabling automatic TRIM in the OS, and running a daily or weekly batch trim instead. Not sure if there's an equivalent to this in Windows.)
Given the way Windows 8 and 8.1 carry out TRIM I'd say Windows is already doing exactly that. Defrag.exe controls TRIM now and is scheduled for weekly use.
just brew it! wrote:Ryu Connor wrote:just brew it! wrote:(I've seen a number of Linux users recommend disabling automatic TRIM in the OS, and running a daily or weekly batch trim instead. Not sure if there's an equivalent to this in Windows.)
Given the way Windows 8 and 8.1 carry out TRIM I'd say Windows is already doing exactly that. Defrag.exe controls TRIM now and is scheduled for weekly use.
That makes a lot of sense, provided what it does is just issue TRIMs for unallocated space instead of actually trying to defragment.