There are many
questions confronting today's managers about the state of small form factor HDDs
and flash memory:
- Will solid-state storage
replace all rotating media over the long term?
- Where does flash threaten rotating storage and where does it not?
- What applications can take advantage of flash, and what applications
cannot?
- How will media's predictable price drops change our work/play habits?
Although digital cameras have solidly moved from rotating storage to flash, and
although USB Flash Drives,
the little keychain fobs that store hundreds of megabytes of information, have
already displaced floppies
on many PCs, there is no reason to anticipate an exodus away from today's
optical and
magnetic storage in PCs
and other traditional applications.
There are lots of angles that designers consider when choosing flash
versus HDDs. Price per megabyte is often discussed, but this must be weighed
against total system cost, which is an equally important factor, as we'll see
soon. In many applications physical size and weight limit the use of HDDs
(although today's form factors get away from that in very many cases).
Energy consumption
and robustness are arguments often used by purveyors of solid-state drives, but
these factors are important to a more elite audience, so they will not be
discussed in this article. But one very important factor is file size, which we
will see counterbalances costs to become one of the top reasons for selecting
one or the other media type.
Cost is a Key Factor
The single most important factor in any consumer system is cost.
As long as the desirable features can be brought to the market for an affordable
price, the product should succeed. The big trade-off is cost vs. features. It
would be nice if the world would stand still, and a tidy answer to the question
of costs could be derived. This is not the case, yet the dynamics of HDD and
flash pricing are clear enough that we can draw some important conclusions that
we can use to foretell the future.
The following chart illustrates historical price per megabyte trends
of flash, DRAM, and HDD. We show
DRAM price per MB because
flash is a new enough technology that we only have eight years' worth of data.
Since both are semiconductor memory technologies, both are driven by the same
factors, so both can be expected to follow the same trend line.
|
The chart is plotted using a
logarithmic vertical axis. This helps us to read the chart because steady
growth is drawn as a straight line on a semi logarithmic format. Straight lines
have been overlaid on top of the data to illustrate how steady the trends really
are.
Flash prices can be expected to continue to drop at an annual
rate of 30-40%. DRAM has followed a
32% average annual price decline for over 30 years. The wavy portions
of the line indicate price flattening during shortages. The one place where
DRAM prices drop significantly below trend occurred during a period of prolonged
below-cost pricing. As a result of this policy foreign competitors' wrists were
slapped by the US and European governments. The portions of the curve that rise
above the trend line indicate profits during shortages - nobody can begrudge a
memory manufacturer taking profits by holding prices firm during a shortage.
HDDs underwent a change in the late 1990s, where the shallow price
decline, which was on a collision course with semiconductor memory, suddenly
became much steeper. Semico does not closely track the HDD market, but we have
been told that this is where the GMR head started to be used. The new slope
accelerates HDD's price per megabyte decline to a point where it looks like
semiconductor memories will not be able to catch up. This chart indicates that
HDD is priced three orders of magnitude lower than flash, but the flash data is
an aggregate over all densities of chip - the higher density chips used for mass
storage are about 1/30th of this price, while the HDD line is based upon the
least expensive version. Additionally, the small form factor HDDs that are used
in portable applications are several times as expensive as the HDD trend line
would indicate. All in all there are only about two orders of magnitude between
the average mass-storage flash and the average HDD.
For the sake of simplicity, let's not assume that there will be any
more breakpoints in either curve (which is an adequate assumption for the near
term) and let's look at these two price declines as if they are roughly
parallel. These two assumptions will help develop arguments that can be refined
by the reader at a later point.
With the kind of price drops the preceding figure illustrates, where
does flash threaten rotating storage and where does it not? At first flush it
appears that there is no contest. Why is anyone currently using flash? The
answer requires for us to look at the big picture, to understand where pricing
is at a system level. The following chart compares flash prices against HDD
prices across a wide range of capacities for a set point in time. The numbers
are approximations made in 2004, and the reader may want to plug in more
relevant numbers. We argue that this won't change the nature of the chart that
much, just the placement of the breakpoints. This chart was developed to
compare a $100 64Gb HDD to a $10 32MB flash card.
|
At the point where the two
cross over the price of that much flash has become equal to the floor price of
the HDD. For the numbers used here that occurs with slightly more than 512MB of
flash. At this juncture HDD and flash add the same cost to the system, yet HDD
offers the designer 100 times as much storage.
Looking farther away
from the origin we see a point where the HDD price starts to rise. This is the
capacity at which multiple drives must be used, so pricing suddenly becomes
proportional to storage requirements. An example of such an application would
be a RAID array or a collection of blade servers. The HDD and flash lines track
from this point on, but the flash line is over 100 times the price of the HDD.
There are few applications that would gravitate towards flash at such an
expense, but this is the realm of solid-state drives, which offer faster seek
times, extreme mechanical ruggedness, and low power consumption at a
considerable premium.
To boil this down, we see that for very small storage requirements
flash gives a lower overall system cost. Once flash approaches the floor price
of the HDD, then HDD wins hands-down from a pricing standpoint, and any decision
to use flash would have to include more considerations than simply cost.
Combining the Charts
What do the two charts show us together? We were unable to come
up with a reasonable representation of this in a picture, so we'll have to
describe it in words.
First, let's assume that HDD and flash price per megabyte will
continue behave as they do in the first chart and track each other's decline
over time. Next, let's assume that the floor price remains intact (which is
roughly the case for any given form factor of HDD, although the price will not
necessarily be $100) then we find that the shape of the second chart remains the
same over time, yet the two lines move to the right at a rate of one doubling in
capacity about every year.
The net effect here is that flash will
become the better choice for larger and larger applications over time. HDD will
continue to be the less expensive choice for applications where flash's price
equals or exceeds the floor price of an HDD.
This means that applications that need an HDD today may be able to cut
costs by going to flash in future years if the file sizes don't expand over
time.
So now we have the question: "How much capacity is needed by my
application?"
File Sizes
File size is the final key factor in this
equation - what is the size of all of the files that must be stored upon the
media?
Certain files will not grow over time. A good example of this sort of
file is an MP3 song (approx. 1.5MB for a 3-minute song). Feature films are also
unlikely to grow much in size. Today's 90-minute feature film consumes about
2GB of storage. Arguably HDTV will double this, but this is not an ongoing
increase - it is a one-time event unlikely to be replicated for over a decade.
Another one-time event that works in the opposite direction is the migration of
videos from the MPEG-2 compression standard to MPEG-4, a change that is likely
to cut the storage requirements of a film to as little as a tenth of its MPEG-2
size.
What about camera megapixel requirements? Many in the digital camera
business believe that camera megapixel offerings are about to stagnate at a
certain level (although there is no agreement if this is 3 megapixels, 5
megapixels, something between these two, or something larger). What Semico
knows for a fact is that while HDD and flash capacities double approximately
every year, camera megapixels have historically grown at a much smaller rate of
about 19% per year. For the sake of argument we will assume that camera
megapixels will continue to grow at this very low rate.
Other files will grow as fast as or faster than the growth of the
capacity of the media. A prime example of this is software, whose size seems
only to be limited by the maximum practical HDD size in the system. The data
files that go with this software are equally unconstrained in growth.
For those applications that don't grow and for those that grow very
slowly, the decline in cost in the first chart implies that the amount of
storage in the system (a camera, a TiVo, an MP3 player, or what-have-you) will
increase steadily allowing users to store an ever-increasing number of
fixed-length files in the system. Napster's recent advertising campaign in the
US points out that you could spend $10,000 to fill your iPod to capacity with
songs. Not only are users unlikely to embrace this idea, but insurance carriers
are even less likely to offer replacement value for the loss of such a
well-filled iPod. Even if songs cost nothing, how much data is manageable
through the interface of a portable device?
Here's another similar example: By Semico's estimate the average
capacity flash card in 2008 will be able to store almost 1,000 photos taken by
the average camera purchased that year. We wonder how likely it will be that
users will wish to manage 1,000 photos through the camera's interface.
It is reasonable to expect users to max out on the number of these relatively
fixed-size files that they wish to manage on a portable device. Although we
don't know what this number is, we believe that the maximum manageable number of
fixed-size files a user can deal with will limit the memory size desired by the
user. In turn, this should cause systems to eventually migrate from HDD to
flash.
This sounds very negative from the viewpoint of the HDD maker, but new
applications promise to come to the rescue.
New Applications Will Change How We Work & Play
Just as the TiVo is changing how people watch television, and as
digital media is changing how people take and share photos, the availability of
the right kind of storage will change other aspects of our lives.
Video consumes prodigious amounts of storage. TiVo and other personal
media recorders will continue to consume larger and larger HDDs for years to
come, never once entering into a realm where flash can compete. Many people
will find themselves with a stash of hundreds of hours of un-watched video,
waiting for that "some day" when they can get caught up.
One application that Semico expects to emerge from ever-decreasing
media prices will be camcorders that are based either upon HDDs or flash cards.
Many examples are on the market today, but either they are either too expensive
(in the case of HDD and quality flash camcorders) or they offer insufficient
quality (the other flash camcorders, which record low-resolution images.) Once
we have reached the magical juncture of quality and price, the market should
open up and perhaps explode.
We are sure that other similar
applications will emerge, and that creative minds will imagine ways to use
growing HDD capacity that we have never even dreamed of.
What to Expect Going Forward
Here are some rules of thumb that will help with any decisions about
when flash will appeal and when HDD will be the better choice:
- Flash will be the leading choice in portable applications where a limited
number of small files are used because it will offer the lowest overall system
cost.
- Other limited-capacity applications will also gravitate towards flash.
- In applications where file size or the total number of files to be stored
is of more concern than total system cost, HDDs will prevail.
- Everything will change over time, as declining prices cause flash to
replace HDD in portable applications where storage requirements reach some
natural limit.
- New applications will emerge to take advantage of lower prices afforded by
HDDs.
...Semico
Research profile, flash,
HDDs
...Later:-
2008 - Sun's
CEO
Jonathan Schwartz,
cited the above article in his blog
Anything
But a Flash in the Pan
See also:-
the top 50 SSD
articles on StorageSearch.com - updated monthly
STORAGEsearch is published by
ACSL |
