SSD mounting?

[x88x]

I've seen some product claims of over 2 million write cycles

Those would be drives using SLC chips. They're lower density but higher speed and have much longer lives. Good ones do in fact have available write cycles in the millions (usually around 1-2, from what I've seen). This is why whenever you see a company marketing their SSDs to enterprise customers, they almost always use SLC chips.

[Konrad]

Interesting data.

The failover chips are a fine idea, insurance against diminished storage capacity. I wonder if they're actual discrete chips or just inactive blocks in the normal chips which are reserved for this purpose. Either way, activating new chips to replace failed ones can't bring back lost data, eh?

But I suspect your calculations aren't quite right, sorry. Your example of "rewriting the entire drive four times a day" seems too simplified, I think it's more like "rewriting the blocks on the drive where the file system is stored many hundreds or thousands of times a day" (not quite as severe if wear-levelling moves these blocks around for each write operation, though still pretty bad since changes to the file structure usually also involves simultaneously writing the files themselves elsewhere on the drive as well - exceptions can occur when files are moved and always occur when files are just deleted or renamed ... then again, maybe this overcomplication basically leads to the same results as your logic). RAM buffers in the SSD controller can cache a large number of cumulative changes before committing them all in a single write operation, which no doubt adds a lot of longevity. (Kinda strange, when you consider that RAM buffers in HDD controllers are used to minimize excessive read operations instead.)

Are SSDs (or at least some SSD products, like the SLC versions you mention) already being considered reliable alternatives to HDD storage within the server/enterprise world for mission-critical machines?

[x88x]

The failover chips are a fine idea, insurance against diminished storage capacity. I wonder if they're actual discrete chips or just inactive blocks in the normal chips which are reserved for this purpose.

It depends on the drive. Most manufacturers do inactive blocks, but I think several do whole separate chips for their higher end drives.

Either way, activating new chips to replace failed ones can't bring back lost data, eh?

That's another point where the controller comes into play. With most modern SSDs, no chip or block will ever actually completely fail under normal operations. Instead, when the controller sees that a block is reaching its maximum write operations, it moves the data in that block to an unused block that is not nearing its max write operations.

RAM buffers in the SSD controller can cache a large number of cumulative changes before committing them all in a single write operation, which no doubt adds a lot of longevity.

I believe this is the case, yes. Between the RAM buffers in newer and higher-end SSDs, wear-leveling, and optimizations in modern OSs to minimize write operations in order to be more SSD-friendly, the wear on any one chip is increasingly minimized.

Are SSDs (or at least some SSD products, like the SLC versions you mention) already being considered reliable alternatives to HDD storage within the server/enterprise world for mission-critical machines?

I can't speak for everyone, but I know where I work there are talks to start implementing SSDs in production. But then, we're also in the process of replacing all our tape backups with HDDs, so maybe we're not exactly industry-standard. I know there are at least several SSD manufacturers who cater exclusively to enterprise and military customers. pureSilicon is one, OCZ's 'e' designation drives are specifically marketed toward enterprise customers (they even have an MLC SAS drive that they're marketing towards enterprise customers). The pureSilicon drives only have an advertised MTBF of 2 million hours, but the enterprise OCZ drives have an advertised MTBF of 10 million hours. Now, obviously neither of these can be empirically proven (even 'just' 2 million hours is over 228 years), but if they're apparently willing to put their reputation on the line for that, and even if they only last 1/100th of that time (still over a decade for the OCZ drives), I'd be happy.

[Konrad]

I'll do a little reading, but I think you've sold me. The next drive I purchase shall be SSD. That might be a little while since I'm floating around 4-6TB of unused HDD and a couple stacks of writable DVDs.