Probably 80% of your hard drive is stuff where the access speed is irrelevant -- movies, for example, since your capability to read from the disk far outstrips your eyes' capability to watch the movie. Photos, Office docs, email, ditto ditto ditto.
Then there are programs -- or even a subset of programs, really -- that actually have appreciable startup times. Office, Eclipse, etc, I'm looking at you.
You could put those programs on the "fast" disk (along with most of the OS, presumably) then make it as big as you pleased with cheap spinning platters. With a little bit of software trickery, you could present the two disks as one physical drive to the operating system (or to the end user) and shift data between them using some sort of caching policy (LRU, whatever).
It sounds like a sexy idea for servers too -- can't afford to keep the entire working set in RAM? No problem -- back up the RAM with solid state and only write to spinning magnetic media when you need long-term non-volatile storage.
Have you seen the startup Woz has started working with?
I was amazed at really seeing the HD bottleneck minimized this way. I always knew about it, but -- wow.
Everybody thinks, "duh, RAM (or SSD or w/e) is faster than disk" but sometimes these sorts of real-world tests still amaze.
And I daresay that in three or four years, this will indeed be considered "normal performance", if SSDs continue to develop.
All hail the SSDs! Maybe in a couple of years we actually will be able to feed our octo-cores some data to work on!
Why should you stop hearing it? Hardware is getting faster and SW is using the capacity.
> If hard drives had been advancing, errr, at all,
They have been, just not uniformly. There have been huge increases in capacity. Less so in transfer rate, but still significant. Seek time hasn't been improving much and rotational latency is stuck.
In the same time, arithmetic and logical operations have essentially become free. Branch mispredictions are a big deal. Cache misses are a killer but memory bandwidth is about to become the biggest problem. (It already is for some people.)
Imagine a WYSIWYG editor that is actually 2^10 times faster than those made 15 years ago.
I think uTorrent has proven small can be loved.
However, given that a 64GB Samsung SSD drive goes for about $500, 24 of them (plus RAID card for that many) is still looking quit steep!
The IT guy from the video is a poster there, so there is some useful information if anyone is interested.
Vista has a feature called 'superfetch' which pre-caches applications in memory so they're never actually loaded from disk. Also, it's hard to tell, but if this computer has sufficient RAM, you could conceivable load everything directly from memory.
Also, the fragmentation test was a little dubious. A computer that new would have very little file fragmentation, so of course defragging would be fast.
That said, it was an entertaining video!
Fragmentation test... well, it is a marketing video, after all. The real question is "why defragment an SSD at all?"
ans yes, i totally agree (re Superfetch)... even with XP, after loading an app once, windows was able to fetch the data again much faster the second time. They should have showed the tests from power to boot to tests.
More exposure, more familiarity with the hardware and advantages, more potential buyers.
So they may have sold an SSD or two.
But then again, I'd probably go Intel and not Samsung.
Still. They got people talking and the video probably wasn't all to expensive to produce, so they got at least SOME value for their money.
Has anyone done performance tests with MySQL on low-latency storage devices? As another commenter suggested for the desktop, strategies of using solid-state for latency bound activities and traditional disks for throughput intensive stuff might be an interesting hybrid to explore.
HDD: http://encosia.com/photos/things/drive-benchmark-10k.png
SSD: http://encosia.com/photos/things/drive-benchmark-SSD.png
Regular (cheap) MLC-based SSDs are good only at reading big sequential files and lose out to 7200rpm drives when it comes to many small writes.
I can't find a ling to a guy who blogged about his disappointing experience compiling C/C++ sources on a regular MLC drive, but I googled Linus Torvald's experience with Intel:
http://torvalds-family.blogspot.com/2008/10/so-i-got-one-of-...
I personally find that I just have all my software open all the time, most of what I need gets pre-cached in 4GB of RAM I have, and I never reboot, i.e. no need to restart anything [unless, of course, freaking Adobe Flash freezes my Mac]
http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=34...
Gives a lot of detail about the various types of SSDs and why the intel is currently the best bet. Warning, reading it might cause you to go into a "gear lust trance" and spend a lot of money.
The ones I've tested are faster than a HDD (and faster than the SSD in a eee PC). They're also cheap.
Bottleneck would shift to USB bandwidth (60 MByte/s) and the PC itself.
This part impressed me most.
I tried opening OpenOffice after watching this video and found that OpenOffice writer opens in less than 1 second and opening all Office applications except for database and impress takes less than three seconds 3.
The difference between SSDs and HDs is definitely noticeable.
So - if seek time wasn't a big factor, a fast conventional drive could be faster? And if you have a very high level of striping (ie. 24 drives), seek time might be significantly mitigated.
Yet, here we see that with a modern system that is mostly only special in the I/O department that Microsoft Office as a whole opens in .5 seconds, which, believe me, is radically faster than Office 95 opened back in the day it was still cutting edge.
Yes, it was a powerful system, but the only thing that was radically more powerful that what anybody can buy for cheap was the I/O system.
Maybe it's "bloat" in the abstract if programs are simply "larger" than they used to be, but really, who cares about bloat if hardware is actually outpacing software development... in everything but the I/O hardware.
I'm just a bit tired of people babbling on about software bloat when this pretty conclusively proves its the I/O subsystem that is the problem.
This proves nothing of the sort. (Office95 was bloated too....)
The time to boot is large because software is huge, scattered all over the disk, and has unnecessary dependencies.
It may be convenient to believe that everything fits in L1 cache, but doing so often results in slow programs. The same principle applies here. Some operations cost more than others. Software which uses slow operations will run slower than software that doesn't.
To put it another way, big O is the start, not the end.
To put those ratings in context, consider that a Western Digital VelociRaptor has a 160 year MTBF.
I'm sure it will happen in a few years. I wonder if there's some starup opportunity here: are there any ways to take advantage of fast, reliable, cheap, flash-like storage becoming ubiquitous?
Probably you could add more fine-grained extension hooks to programs without them becoming bloated, by having plugins loaded lazily at run-time from a solid state drive. I'm convinced that the next big thing will be software that quietly integrates with other software, so you don't even have to click on an icon to use it.
