Comment by jrockway | Hacker News Reader

Comment by jrockway | Hacker News Reader

Seems pointless; the flash only works as a read cache. I already have a 4G read cache that's much faster, it's called main memory.

SSDs are nice for lots of random reads/writes; I noticed the biggest improvements on things like "apt-get upgrade"; deleting files, reading tarballs, updating a database. This is where SSDs excel, but the SSD that's part of this disk won't help with any of that. When you download the tarball, it hits the magnetic disk. Then when you extract that, you read from the disk, and write to the disk. Then you read the package database... all from disk. You update the package database... yup, disk write. And so on; this drive will do nothing to help this use case. Anything that could be cached will be invalidated by a write. So you get no speedup on the only workload other than booting that helps a regular user.

If you're thinking of buying this, buy a regular disk instead, and save the $60 premium for a real SSD. Even Intel's "budget" 40G SSD will make your computer noticeably faster... but this thing won't.

Main memory doesn't persist across reboots; this does. Hence, potential increase in boot speed for the main OS, as well as a decrease in start-up time for an application the first time you run it after starting the OS.

And yep, no increase in that use case. Faster in the use case of opening programs (as mentioned in the article), faster in the use case of opening small files, and so forth.

Your last sentence is needless hyperbole, and is contradicted by the last sentence in your middle paragraph. Not that I disagree with your conclusion, though.

I boot my machine maybe once or twice a month. All other times, it wakes up from sleep or hibernate (which this drive won't help, of course). Most people, especially mac users, are the same way. Rebooting died with the 90s.

As for starting apps, they start when I log in, and most initialization is CPU-bound, not IO-bound. If you do it differently than this, a software or workflow fix seems like a better idea than a very expensive drive.

True enough... but you're underestimating the subjective performance increase. It's astounding (not with this drive, but with a full SSD)

Since switching my mac to a decently quick SSD and doing a few other tweaks to go along with it - I actually shut it down for the night or when I know I won't be using it for a while - because it shuts down in less than 2 seconds and boots to fully-usable in 20 seconds - and apps still start instantly, suddenly rebooting is no longer a thing to work around or avoid. I don't worry about keeping anything running to speed things up, because everything is just that fast.

Plus I can just yank it off the table without waiting for drive spindown if I have to run off to another meeting.

> Main memory doesn't persist across reboots; this does. Hence, potential increase in boot speed for the main OS, as well as a decrease in start-up time for an application the first time you run it after starting the OS.

So use system daemons like 'preload' or 'readahead'.

And come to think of it, a daemon can be upgraded and the heuristics & statistics improved. (Even if one can upgrade the hybrid drive's firmware, I'm not sure I would dare do so!)

Yeah, exactly. The reason boot times are long with rotating disks is because our software is really really dumb. Smart software would make "real" boots as fast as un-hibernating. Same for app startup. (/me contemplates cryopid-ing every app that takes a long time to startup, putting the data in a file that is guaranteed to be a single read operation, and enjoying fast speed without a SSD. But I have a SSD already, so meh.)

So set those daemons up on Windows.

(Linux, sadly, isn't necessarily the intended market for things like this.)

Rebooting is rare compared to regular use.

The box I'm writing this on has an uptime of 148 days, the last time it went down was to put a second graphic card in it. I really don't see how a 'faster boot' would be of any benefit, in fact any benefit in faster boot time would have to be balanced with how long it took you to install that drive, format it, install your OS and so on in the first time.

Use cases should center around that which happens frequently, optimizing boot time for anything other than a netbook or a laptop that you don't 'sleep' is a case of premature optimization.

For all the other use-cases you sketch main memory is much more effective than a cache in a device at the other end serial link.

> ... for anything other than a _netbook or a laptop_ that you don't 'sleep' is a case of premature optimization.

From the article: "Take a standard issue 7200 rpm, 500 GB _notebook SATA drive_. Add 4 GB of fast and reliable single-level cell (SLC) flash."

(Emphasis mine in both cases).

And you don't use sleep ?

Besides that, it seems that the boot speed increase according to the article is next to non existent, and the comments point out that the application starts were actually done without wiping the OS cache.

Thanks for sharing your personal computing habits with us. I happen to reboot quite often, switching between Ubuntu and Windows. I tried using a VM, but there were too many issues. With an SSD, rebooting from one OS to the other is fast enough to not be annoying, so I can have my dual-boot cake and eat it too. I have paid, and will continue to pay good money for this.

You should try a real SSD then.

I read the description differently. The article says, "Put those blocks in the flash ... leaving big sequential I/Os to the disk." Sounds to me like the small blocks that are accessed (read or written) often are physically located in the flash---not just cached. If your package database isn't sequential on disk and it's accessed a lot by your system, it will probably be located in the flash.

Reply on news.ycombinator.com