Quote: Original post by samoth
While this is certainly possible, registry and the swap file are SSD nightmares.

This is because, and not to mince words, current generation 'affordable' SSD drives are shit.

AnandTech recently did a review of Intel's SSD drives, during the process of which they hit upon a problem with current drives which use a jmicron controller.

With even a 10% randomness to the writes the drives performance goes to hell, to the extent of seeing pauses while buffers block because they are full.

Intel's SSDs on the other hand don't have this problem at all.

The other issue I believe you hinted at was life time, I can't speak for the current drives but again Intel's won't have a problem; they have garrenteed that you can write 100G/day to these drives for 5 years before they fail.

SSD are also slightly cleverer than your normal magnetic drives in that the controller on board knows when it's going to fail, as such if it does fail it fails BEFORE a write operation, leaving all your data intact and just needing to be moved to another drive. In theory the drive software could even give you a slight warning to let you know things are about to go south, though in practise this probably doesn't happen.

ssd's are quite affordable if you look at them differently:

i've now bought a 450$ ssd for my notebook (32gb slc memory, mtron 3000) as it's the only fast one able to connect to my notebook.

now that's not cheap. then again, no high-end gpu is. no new pc/notebook is. i've boosted my one year old notebook for 450$ to a complete new performance level (up to 30times as fast than before in scenarios i care (opening firefox .. :)).

for that money i've done the best i can: solved my bottleneck completely. the only other way to spend it is to buy a new notebook, that would cost much more, and would NOT solve the bottleneck (except if i buy a notebook including ssd, which, again, costs even more).

ssd's are in no way cheap (good ones). but if you have an aging notebook, thinking of getting a new one because of it. think again. for half the money, you'll make your current one much faster than your new one.

then, it's not soo expensive anymore :)

if my notebook would've had an s-ata adapter, i'd get the intel ssd.

btw, when an ssd fails (and the controller is not stupid and craps then out to make the whole ssd inaccessible), you can't _write_ to the ssd anymore. reading is still supported.

so a failed ssd is a non-issue for your data (if the vendor implemented it right). a failed hd is about never recoverable once you get the click-noise :)

as i have an slc memory non jmicron based ssd, i don't have turned of anything from vista (yep vista on a 32gb disk, i'm that crazy :)). except system-restore as i have a home-server for backup. it just takes up huge amount of space :)

Quote: Original post by phantom
This is because, and not to mince words, current generation 'affordable' SSD drives are shit.

Well, yes, for write-many applications they are. But again, for data you write once and read many times, they really rock. I have one of the OCZ Core drives, which are pretty much the cheapest ones on the market, and I'm very happy with it. Having development headers and compiler toolchain on the SSD reduces overall compile time on big project on the order of minutes (!).

Quote: Original post by phantom
The other issue I believe you hinted at was life time, I can't speak for the current drives but again Intel's won't have a problem; they have garrenteed that you can write 100G/day to these drives for 5 years before they fail.

Yes, though I'd be careful with that estimate. It's based on "if you write 100 GB, that's so and so many blocks, and the disk has so and so many blocks total, ... write levelling, blah blah.... makes so and so many years".
However, what's not in that calculation is that in everyday usage, you don't have writes the size of an ISO image, but much smaller ones. And worse, invisible ones.
For example, fire up filemon and look at the frightening number of SET_ATTRIBUTE calls if you do anything in Windows Explorer, even if you only open a folder. Most antivirus software will add its own secret tampering on top of that, each time you access a file (Kaspersky, for example, generates an insane amount of disk writes behind your back).
If you have 100 such "micro-writes" of only 4 or 8 bytes maybe, they are nevertheless seen as 100 times 512 kB by the SSD. So... browsing a few folders may be seen as 50 MB of writes by your drive. Hopefully the file system cache will consolidate some of these writes, but who knows.

(crap antivirus software breaks your SSD rant)

So now I have to rectify my own post, in defense of antivirus software, with what I just found out: The by far worst ever offender is Seagate Replica (also known as Rebit).

Having had the second pure read-only SSD break after only 2 years, I wondered how this could happen. Copying some program files to a drive once and only ever reading from that drive (maybe upgrading a program to a new version once every 2 months) should be no challenge at all. So what the heck is wrong?

Running Procmon (in a configuration that does not filter out writes to ntfs/system files), it turns out that Replica does 5 isolated, sector-sized or smaller, unbuffered writes per second to the drive, for as long as the machine is up and running. Even when it doesn't have anything to do because the backup disk isn't attached at all. That's 18,000 write cycles per hour for nothing. Bang, you're dead.

Hybrid doesn't make sense to me. You can just use ReadyBoost if you want an increase in performance at startup and program load times. It works with any flash drive, at a much cheaper cost than a hybrid drive.