People keep telling me that I should have an SSD in my laptop… I’ve seen some hints from people I trust that one might prefer to buy Intel SSD. Now I wonder whether they were referring to any Intel SSD, or only to the enterprise drives with Intel controllers.
As I ponder this, I like the idea of checksums on data (btrfs). And then maybe a second spinning rust drive for larger files and as a backup target for the SSD. Someday perhaps I’d use btrfs for both and send filesystem snapshots to the second drive, but right now rdiff-backup is my trusty backup sidekick.
H. Peter Anvin February 26, 2014 23:36
H. Peter Anvin February 27, 2014 00:12
(Obvious disclaimer included.)
Eugene Crosser February 27, 2014 00:24
I’d say power loss in a notebook is rather rare, I’d not give resilience to power loss very high priority.
I have some run-of-the-mill 250Gb Intel SSD in my notebook. I have no problems with it, but I don’t stress it too much. (I have PCI OCZ Revodrive in the desktop, which I do not recommend.)
Michael K Johnson February 27, 2014 06:35
I notice browsing ark.intel.com that apparently I need to spend around $1K these days (for a drive with any real capacity) to get supercapacitors in the drive (if I correctly interpret “Enhanced Power Loss Data Protection”). The 530 has “End-to-End Data Protection” but the “help text” doesn’t say anything meaningful; it’s an “i++ // increment i by one” style description. My google foo seems weak; the non-authoritative descriptions on teh intarwebs imply that it’s nothing more than that the drive has error correction and checksums. Duh?
The drives with “Enhanced Power Loss Data Protection” use lots more power than I would want in a laptop, even if I wanted to spend an order of magnitude more than on spinning rust for a primary drive. It seems reasonable to draw lots more power to get lots more performance so I’m not complaining about that per se.
I’m thinking spinning rust in the ultrabay holding frequent rdiff-backup incrementals, which I can easily remove temporarily when I need more battery life, and btrfs with checksums on a consumer-level SSD, should make it possible to recover from write failures. Maybe with a
find . -mount -type f | xargs cat > /dev/null on each filesystem after a power loss, and occasionally as the equivalent of a “surface scan” for all the “surface” that matters, in order to make sure to notice checksum failures.
David Megginson February 27, 2014 07:29
How would an SSD behave on a laptop doing a lot of server-type development (e.g. running a SQL database, Apache HTTP server, lots of git pulls, etc. etc.). Would continuous small disk writes slow it down or wear it out sooner?
Curtis Olson February 27, 2014 08:09
You guys are discussing SSD’s at level much deeper than I can participate in, but let me just say that my linux computers that have SSD root drives boot 10x faster than they did before with the spinning drives. One of my Fedora machines is setup for autologin and I haven’t exactly timed it, but I’m pretty sure I see my desktop pop up < 5 seconds after the machine finishes POST-ing. None of my SSD drives have failed or worn out yet. My opinions may be subject to change as more data comes in, but so far I’m pretty happy with them. For whatever it’s worth, I’ve got my root setup on the SSD drive and /home and other data on 1 or 2 Tb drives – cost is still a big driver in my SSD purchasing decisions.
Alan Cox February 27, 2014 16:03
SSDs are awesome.. until they are not.
Disks at least usually show some hint they might be about to fry, and generally have the decency to fail cleanly rather than silently and quietly corrupting your data.
H. Peter Anvin February 27, 2014 17:00
Unless they don’t. I’ve had a bunch of disks die completely without warning, which usually means electronics rather than mechanics killed it.
Michael K Johnson February 27, 2014 22:17
+David Megginson Generally, continuous small disk writes wear it out sooner. How afraid are you of silent corruption? I’m afraid of it, so I think I’ll use btrfs.
The Intel drives have five year warranties, but now I’m reading the fine print. If you buy the “OEM” version of the drive (“Reseller kit”), your warranty “ends on the SHORTER OF the specified warranty term or the date when the media wear-out Indicator value for the drive reads 1.” For a bare drive, it’s five years with a mere recommendation that you back up or replace when the wear indicator approaches 1, which is why the bare drive costs more than the reseller kit, I guess. Of course that covers the drive, not the data! So if you’re going to do development on it, go for the version that doesn’t have a wear-limited warranty. It’s not that much more expensive…
I hear that postgresql is much friendlier than mysql to SSD. I have no data to back that up.
+Curtis Olson I really don’t care much about booting slightly faster. It’s /home I want to go faster.
I have the usual geek pile of dead spinning rust, but in my case I am reasonably certain that I didn’t see massive silent corruption. And the spinning rust pile gets smaller when you add children and a toolkit! Magnets!
Michael K Johnson February 27, 2014 22:21
I think I got that backwards. The “OEM” drive that is wear-limited is the bare drive, and the “Reseller kit” that costs about $20 less for (otherwise) the same drive seems to be the one with the non-wear-limited warranty. I’m going to triple-check that against the warranty documents from Intel before I click “purchase”…
Cristian Gafton February 28, 2014 03:58
The Samsung Pros are also pretty damn good. I’d say they are even better for a laptop due to much lower power consumption - both when idle and when busy - compared to Intel Enterprise ones.
(I assume you’re looking at Intel for their 730 or 3700 series and you give the proper disconsideration to any of their SandForce junk)
Michael K Johnson February 28, 2014 06:51
No, I was not looking at a 730/3700. Reviews seemed to me to be saying that Intel implemented SF more conservatively; lower peak performance but consistent and not failing. Not, for example, like OCZ reportedly deciding to ignore SF and flip every bit they could for performance at the expense of data integrity.
Lost in a maze of twisty web reviews, all alike, I previously missed that the samsung pro drives are 2bpc. I was not interested in becoming a test consumer for 3bpc flash… Looking at warranty terms, even the “PRO” drive warranty says “for the shorter of: (i) a period of five (5) years for the 840 PRO Series and three (3) years for the 840 Series and 840 EVO Series, (ii) the period ending on the date when the SSD has exceeded its TBW (Total Bytes Written) threshold as may be indicated by Samsung’s Magician Software” — and the Samsung site is 100% useless for finding what that limitation might actually be. That’s hardly “industry-leading” or “generous” as Samsung claims. “You used your drive, ha ha, sucks to be you!” Even the “Technical Specifications” on their “Data Sheet” (half full of marketing cruft) says absolutely nothing about total write capacity. It’s only when you read the actual terms of the warranty that you find out that they will cut you off at some place that they refuse to document in text. That doesn’t exactly inspire confidence! Nor does the fact that the link to the manuals in the datasheet doesn’t provide manuals. sigh
I’m not anti-Samsung. I have an S3 (work) and S4 (personal) and I use samsung SD cards in my cameras.
Cristian Gafton February 28, 2014 08:40
Michael you are guaranteed not to exceed the TBW threshold during regular laptop use for a good bit more than the five years they offer a warranty for. I use mine quite heavily, and their Magician tool tells me I have exhausted 1.46% of my drive’s write life in the 28 months I have had it - total written data ~21TB. You’ll be fine, stop obessing over this stuff.
Michael K Johnson February 28, 2014 11:22
Thank you for the information. My complaint is that Samsung should be providing that information. It’s not appropriate for them to have warranty limits that you can’t discover from Samsung without purchasing the device. You’d think it stupid if they advertised the time limit on their devices as something you could find out by running software after you purchased and installed the device. At least, I would.
(Given that write limiting is probably a factor of capacity, I’ll assume that TBW limit on the PRO drives is something around 3PB/TB based on 21TB/0.0146 being near 1.5PB)
Eugene Crosser March 03, 2014 09:37
+Michael K Johnson, I have no idea how useful is this, but in case you can make something of it, here: http://www.theregister.co.uk/2014/03/03/review_intel_730_480gb_ssd/
Michael K Johnson March 03, 2014 16:47
+Eugene Crosser — thanks, worth reading!
Overclocked doesn’t appeal that much to me. Capacitors for write protection does. Price, about 20% higher than the Samsung; now I know why the call them supercapacitors! ;) They aren’t showing up on a search on ark.intel.com yet so I can’t look up power consumption.
Overall, a reason to wait a little longer to see how they compare.
Cristian Gafton March 03, 2014 20:23
Power consumption ain’t that great for mobile use at least according to: http://www.anandtech.com/show/7803/intel-ssd-730-480gb-review/7
“As I mentioned in the first page, the SSD 730 doesn’t support any low-power states and hence idle power consumption comes in quite high at 1.34W. For a desktop this isn’t an issue but it’s clear that the SSD 730 isn’t suitable for mobile use. Load power consumption isn’t as terrible, but it’s still relatively high.”
David Megginson March 03, 2014 20:57
I’d always assumed an SSD would extend my laptop battery life, but I guess not.
Michael K Johnson December 05, 2014 06:44
http://techreport.com/review/27436/the-ssd-endurance-experiment-two-freaking-petabytes confirms both Cristian’s advice to quit freaking out and the choice of the samsung pro. ☺
Imported from Google+ — content and formatting may not be reliable