Slashdot Mirror
The Joy of the Flash Drive
An anonymous reader writes "A post to the C|Net site covers the numerous benefits of flash drives, such as speed, temperature, and battery consumption. The perk author Michael Kanellos is most fond of? The distinct lack of noise. 'The notebook I'm testing--a Dell Latitude D830 with a 64GB flash hard drive from Samsung--hasn't emitted a sound in three days. Flash drives, which store data in NAND flash memory, don't require motors or spinning platters. Thus, there are no whirring mechanical noises. Compare that with my T42 ThinkPad. It sounds like a guinea pig got trapped inside, particularly during the start-up phase. Vzoooot. Cronk, cronk, cronk. Zip, zip. (Pause.) Gurlagurlagurla...zweeee. '"
With this shift to Flash drives for data storage, I wonder if this is good or bad for data archival. With magnetic media, if there is a head crash, at least some data can be recovered. With flash, even though it has no moving parts, if something happens to make a large amount of blocks unreadable, there isn't any real way to recover the lost data.
SSDs are *not* slower than mag platter drives. Get one and try it before firing your mouth off, private!
I hate printers.
You do realize that the speed of the drive isn't limited to the speed of a single chip, right? You know, they can wire the things up in parallel...
Blindingly bright blue lights are no match for the depths of black provided by a Sharpie (tm) brand fine-tip permanent marker. One swipe across and you go from "blindingly bright" to "pretty gosh darn dim." If you're not sure if you want a permanent solution, use a permanent marker on a piece of Scotch (tm) tape and slap that on the LED. Instant dark, but you can still tell when it's glowing.
I've been looking for Flash drives for a while now, and it seems the best option at the moment perfomance wise is the Mtron Pro series at 120Mb/sec. But 32 gigs will cost you 1129.
Which is why I wonder how Dell and Apple and everyone else can provide 64 gig SSD options for their notebooks for less than 1000 dollars. None of the brands had any info on the specs of the drives easily locatable, and I am worried these are the low end SSDs that are much much slower... which is a shame, because performance driven users would probably prefer better drives even for an extra 500 to 1000 dollars.
Later this year Intel is suppose to release 200Mb/sec 80G drives, which is really the only reason I haven't gotten one yet, but I have yet to find any info on pricing.
Flash media is considerably slower than hard drive media at the same price point. This is mainly due to economies of scale: there is a huge demand for low cost, moderately high performance desktop and laptop hard drives, while the demand for flash is for dirt cheap, low performance usb fobs. This is likely to change over time, but it will take years. Production methods for low unit-cost, high performance flash chips have to be developed, fab plants have to be built, all the usual problems.
Flash media (NAND-gate type) is fundamentally slower than hard drives for sustained serial write behaviour, where the seek penalty does not apply. This is not likely to change, since performance for both technologies should increase at roughly the same rate; so long as NAND-gate technology is the best we have, hard drives are still going to be around for those workloads that need that kind of thing (various forms of audio/video work, some database stuff, scientific applications). It's faster for the other major operating modes (all read modes, random-access-write, latency, etcetera), so is likely to give overall better performance for desktop computing workloads. There are experimental technologies in the labs that can outperform hard drives in the sustained serial write mode, but those aren't on the market yet, and may never be. They've been promising us MRAM for twenty years now, and still haven't come up with a product.
Limitations in current flash products mean that everything on the market is also slower than hard drives in the random-access-write mode. That's a problem with a known solution, there just isn't anything on the market that does it yet. This should change in the next generation or two.
I've personally had good experience with a flash-based root file system, but I implemented it with a ide-CF adapter. I picked a bit too slow card for the root, though, but the point of the device is to provide the contents of 4 hard drivers software-raid5'd over the network, so it isn't such a big deal. I don't have swap in the machine.
I've earlier set up a box to boot from a ide-flash device, while the actual root file system was on LVM. It worked nicely too.
Older hard drives were often noisy, and probably some models still are. I had one where R/W sounded like someone trampling broken glass and metal. IIRC, the Seagate Barracuda was my first drive which I couldn't easily hear working.
Well, I think that the noop scheduler/elevator should yield the best performance. It has the least overhead, and the scheduling overhead for deadline and cfq are trade offs for optimizing a read head. The anticipatory scheduler adds a 1ms delay, but very little in the way of processing overhead. I'm just wondering how much performance could be yielded if all the various disk scheduling (read ahead, write behind, delays for queues, read and write starving, etc.) code is going to go the way of the... erm, really dead thing...
If I mod you up, it doesn't necessarily mean I agree with what you've said, sorry.
Disc are quiet *now*... The 10k rpm SCSI disc in my old Ultra 1 sounds like a gas turbine spooling up when I switch it on, and you can hear the heads working in the next room with the door shut!
It depends on the workload. With a 512-byte block size and a 9ms seek time then a hard drive will get 56KB/s for a seek-intensive workload. A good drive with a 4.5ms seek time will get a whooping 112KB/s, or around 222 I/O operations per second. Even a cheap flash drive can get significantly more than this. In terms of sustained linear read speed, expensive flash drives are now around the same speed as cheap hard drives (better than the cheapest ones, but not much). Most workloads consist of a mixture of linear and random reads. Exactly where on this spectrum you lie depends on whether flash or magnetic storage is better for you.
I am TheRaven on Soylent News
/obligatory spelling daemon/ 'Perk' here is a mis-spelling used commonly enough to be accurately understood, yet incorrect nonetheless. 'Perq' is short for 'perquisite' roughly 'for the person' from Latin to mean that which benefits someone, usually used to refer to employee benefits. Shortening it to a 'k' makes absolutely no sense unless you assume that most people only ever hear the shortened version and then write it phonetically. Disclaimer: Not intended as flamebait or unnecessary spelling Nazism.
Apple sells a line of Western Digital external HDDs in their retail stores, and while they are pricier than roll your owns, they are really nice in their lighting scheme. They have a long LED on the front - imagine a 'Cylon style' white LED, but vertical. The light changes intensity, position, etc depending on what the HD is doing. File access - light cycles with access. HD noise itself is pretty minimal, and the fan is virtually silent. Best part, when Time Machine hoses my system and freezes the external HD (which it occasionally does), the light goes full-on and stays on, so I know there is a problem. I think this kind of feedback is great, and it eats up zero system resources.
man vmstat
man iostat
Did you know you can fertilize your lawn with used motor oil?
Well, if you're trying to fall asleep, pink noise isn't bad (it sounds just like a waterfall):
sox -t nul /dev/zero -t ossdsp /dev/dsp synth whitenoise lowpass 100
("-t alsa default" if you're using ALSA)
You assume incorrectly. Back in the 80's there might've been some truth to it. The typical limit today is 1m+ erase cycles (not writes) per erase unit, and a typical large flash units will have many thousand erase units. With a reasonably sized modern unit with proper wear leveling you can write tens of MB/s sustained for more than a decade and still not start seeing failures due to wear. That's without the system keeping a reserve to remap failing units.
Simple.
Open "Task Manager", open "View"->"Select columns". Then choose the columns "Bytes written"/"Bytes read".
You'll be able to spot the offending process without any difficulties.