 |
|
| leading the way to the
new storage frontier |
..... | |
| |
| ... |
by
Zsolt Kerekes,
editor - StorageSearch.com
originally published January 27, 2010 - (for updates see sidebar
articles) |
|
Want to buy some terabytes (or
petabytes) of SSD
storage?
One confusing factor for buyers and specifiers is that
market prices for SSDs have varied more than 100x to 1 for the same
capacity and at the same time! With hundreds of oems active in the SSD market
and thousands of product
news announcements - how
can you decide which SSD prices relate to your own needs? And which don't?
This article will bring a sense of clarity and order into what can seem like
a crazy market.
See also:-
boom bust
cycles in the memory market - an SSD view | |
| . |
the irritating thing about
$$D
pricing!
SSDs are among the most expensive items of computer
hardware many of you will ever buy - with high end models costing more than
high end servers. There's nothing more annoying than spending a large sum of
money on something only to find that someone else you know has just bought the
same thing at a fraction of the price you paid.
Buyers in the SSD
market - who are already confused and irritated enough by the technology
aspects in the SSD shortlisting process - are liable to be stunned by
a new level of random numbers when they look into the issue of pricing.
Why is a terabyte SSD from one company 2x, 5x, 10x or even more than
100x more expensive than another? (And the companies selling those
outrageously costly SSDs keep reporting great business results - so someone
must be buying them - even though other products are much cheaper.) The simple
explanation is - that all SSDs are not the same. And SSDs can do more than
storage. That's why just looking at "capacity" like you would for a
hard drive - does not
give you a true picture of what the product can do - or what it might cost.
a transport analogy for SSD price vs capacity
Outside the SSD market we're already comfortable with
holding seemingly contradictory cost information in our heads - without getting
a headache. Because we know the invisible factors which lie behind apparently
identical purchase decisions.
Let's suppose you need to travel 400
miles for a meeting. Your options are:-
- walk
- bike
- car
- train
- canoe, yacht, raft, speedboat, or river ferry (start and end destinations
have ports on the same river)
- airline (1st class, business class, regular human)
- helicopter
- Airforce One
For the sake of this illustration - the critical distance
you're going to travel is identical. It's 400 miles. But the cost
will vary considerably depending which way you go. And although you may look
at more than one alternative for how to get there - depending what the meeting
is about and your personal resources and preferences - it's unlikely you will
get confused. When it comes to the SSD market - "distance" is like "capacity".
In the travel world - our decision making is simplified by the fact we
filter out a lot of irrelevant choices - which we know from our own experience
are not valid choices.
None of you reading this - are the
President of the USA - so you can instantly filter out the Airforce One
option. It probably never crossed your mind before reading this article.
When
you've learned more about the part of the SSD market -which relates to your
needs - you'll easily be able to filter out confusing price messages. Higher?
Or lower? Who cares? If they're irrelevant you can safely ignore them.
|
 |
Factors which influence SSD Prices
The main factors which influence SSD pricing are listed in the
table below. I've placed them in order of importance - with the most significant
at the top of the list. |
| Factors which most influence SSD
Prices © 2010 StorageSearch.com |
| Speed |
"Speed" is a catch-all term which
includes latency,
random IOPS
and throughput.
Nearly all these factors can be artificially boosted
to look good in benchmarks - and the numbers don't always translate to
application performance due to
halo effects.
Despite the smoke and mirrors, however, experienced users (and this editor) know
the fastest SSDs
when they see them.
In the right circumstances
server users
will buy the fastest SSDs to achieve application acceleration which is
not technically possible without SSDs, or which costs far more - using
additional servers and hard disks.
For the same SSD storage
capacity - street prices for the fastest SSDs can be more than 200x more than
for entry level SSD products.
Before
2007 - the "Year
of SSD Revolutions" the predominant part of an SSD's cost was the
memory type and memory capacity. After that the
SSD controller too
became a significant part of the product price mix. That was the year it
became clear that even within the constraints of using the same interface, and
memory type some designers in the highly competitive
2.5" and
3.5" SSD markets
could use clever architecture
and knowledge of device characteristics to leverage significantly more
performance out of those memory chips. The result was to make their products
more attractive to users - and gave them the ability to charge a higher price.
Those factors - related to SSD IP - had always been true in the
rackmount SSD
market too - but it was in 2007 that it became easier to make like for like
comparisons. |
| Memory |
Here's a simple rule of thumb based on analyzing
published price data - for identical storage capacity - across a wide range of
commercially available SSDs.
- RAM SSDs cost 9x
more than SLC flash SSDs
- SLC flash SSDs cost 2x to 6x more than classic
MLC flash SSDs (2
bits per cell MLC)
Here are some additional price
complicating factors...
- within fast flash SSDs - the amount of
over-provisioning
could mean you've actually got twice as much memory inside the SSD (or half as
much) as you thought.
- new types of nv memory
such as PRAM, MRAM and RRAM etc have appeared in some SSDs - but mostly in
roles as cache
within the SSD - such as Skyera's
rackmounts - rather than as bulk storage (except in experimental devices).
Historically the market ratios between these various SSD memory types
has fluctuated a lot due to demand vs supply, timing of new geometry shrinks,
etc. You can get an idea of the crazy degree of change and direction by
seeing the graph in this classic
RAM vs
flash SSD pricing article. |
| Efficiency |
Some SSDs use nearly twice as much flash - and
also use more expensive related chips such as CPUs to implement SSDs which
are functionally identical to best of breed competitors.
These
differences arise from design architecture, controller IP and software. The
growth of this trend was commented on in the article -
strategic
transitions in the SSD market. |
| Reliability |
Some SSDs have average operational lives which can be 2x, 5x, 10x or even
100x longer than entry level consumer grade SSDs - when they are used in
demanding applications.
Reliability options within the
SSD market - include
internal hardened
data
integrity, enhanced
fault
tolerance or
enhanced
endurance. |
| Interface / Form factor |
Some interfaces and form factors are supported
by more vendors than others.
That means prices may be lower for SSDs
having otherwise similar speed and memory types.
For a complete list
of SSD directories organized by interface type and form factor see the
A to Z list of articles. |
| Security & ruggedness |
There are some other features which can be
important in some SSD applications - but which are not present in all SSDs.
Where they are needed - they can impact system cost (for any given capacity) by
anything from 30% to over 300% compared to other devices with a similar speed.
|
| Ease of installation |
For server apps - in particular - buying the SSD
is just part of the process.
Getting it to work effectively is
another hurdle to cross. I've examined these issues in a separate article -
SSD ASAPs
(Auto-tuning SSD Accelerated Pools of storage). This discusses the situations in
which it's worth paying more for an SSD ASAP - and those others where it's not -
and where human tuning is more likely to give better performance results at a
lower price. | |
Conclusion
SSD
pricing looks complicated - because it is complicated!
It would
be misleading to claim otherwise. There is no such thing as a "single SSD
market". Just as in the transportation analogy used in this article -
there is no such thing as a single way of getting from point "A" to "B".
But you can takes steps to simplify your own SSD price search. A helpful
tactic is to decide which pricing messages to filter in or out - depending
on which features within the SSD cost model are relevant to your own needs.
|
| . |
...Later:-
SNIA has set up a resource
page to help users assess the
cost of ownership
factors for flash SSDs.
It includes a
spreadsheet
which was developed by Intel
and a supporting whitepaper -
SSD
TCO - An In-Depth Analysis of Many Important Factors (pdf) - written by
Esther Spanjer
and Dan Le at SMART
Modular Technologies. |
| . |
...Later:- for
professional buyers and marketers there are lots of
market analysts who
provide guidance on the market price of SSDs.
In the consumer market
a particular specialist in real-time price tracking appears to be
PriceG2. |
| ... |
|
| |
| ... |
 |
| ... |
| how fast can your
SSD run backwards? |
SSDs are complex devices and there's a
lot of mysterious behavior which isn't fully revealed by
benchmarks and
vendor's product datasheets and whitepapers. Underlying all the important
aspects of SSD behavior are
asymmetries
which arise from the intrinsic technologies and architecture inside the SSD.
Which symmetries are most important in an SSD?
That
depends on your application. But knowing that these symmetries exist, what they
are, and judging how your selected SSD compares will give you new insights
into SSD
performance,
cost and
reliability.
There's
no such thing as - the perfect SSD - existing in the market today - but
the SSD symmetry list helps you to understand where any SSD in any memory
technology stands relative to the ideal. |
|
| | |
| ... |
|
|
| ... |
|
| |
... |
 |
. |
what goes into SSD costs?
sugar and spice and all things nice
but
what gets left out counts too. | |
| . |
 |
| . |
SSD history
/ SSD jargon /
SSD endurance -
the forever war
consumer SSD guides
sugaring flash for
the enterprise
custom
SSDs can reduce manufacturing costs
next generation 3D nand
fab yield and the nth layer tax
Exiting the
Astrological Age of Enterprise SSD Pricing |
| . |
| "Jim Handy said the
market is seeing a 10 to 12% pricing drop, Quarter/Quarter, almost 60% since the
year started, in NAND spot pricing which is starting to impact long term
contracts. During supply glut’s like this, DRAM spot prices typically drop
40-60% Q/Q, so maybe there’s more NAND price reductions on the way." |
| GreyBeards
talk FMS18 wrap-up and flash trends with Jim Handy (August 2018) | | |
| . |
|
|
| . |
| See how
much
flash SSD prices
have changed! |
2001 - a
rugged 14GB
3.5" SLC SSD from
Adtron cost $42,000.
2003
- a 21GB 2.5" PATA
SLC SSD from M-Systems
cost "less than $11,000".
2005
- a 10GB 3.5" SCSI
SLC SSD from BiTMICRO
- oem price was $2,999.
2006 (Jan) - a
24GB PATA SLC SSD from Adtron
(commercial temp) cost $5,710.
2006 (Dec) -
Advanced Media
launched a 32GB 2.5" SATA MLC SSD for $1,000.
2008 - a
fast 32GB 2.5" SATA
SLC SSD from Mtron cost
about $399.
2010 -
OCZ said it was
shipping a 32GB 2.5"
MLC
SSD for under $100.
2011 (March)
- Intel said its SSD 320
40GB MLC cost $89 in 1,000 unit quantities.
September 2011 -
the price per terabyte in tier 1 storage arrays is nearly the same for
eMLC SSDs and hard drives - according to
an
article by Jamon Bowen at
Texas Memory Systems.
October 2011 -
Tier 1 FC SAN SSD list prices are about $20K / $30K per usable terabyte
for MLC / SLC respectively (sources
Huawei Symantec and
Violin Memory).
November 2011 -
Coraid said its
EtherFlash solution
(NAS SSD) costs under $10K
/ TB.
June 2012 -
an article in Tech
Report reported the price per GB of consumer SSDs as being mostly
between $1 and $2 across a range of popular models.
August 2012 -
Skyera set a new
low price record for fast enterprise racknmount SSD storage with its
Skyhawk which provides 44TB usable
capacity for $131,000 approx - less than $3K / TB. The new pricing was
enabled by adaptive
R/W DSP flash technology and
big SSD
controller architecture.
September 2012 -
KingSpec said the MSRP
for its 1TB fast PCIe
SSD - MultiCore
model - which has 8x SandForce
controllers on board - and which
started shipping in July
2012 - is under $2,900.
October 2012 -
The cost per GB of consumer
SATA SSDs (64GB to
256GB) has approximately halved in the past year to under $0.50/GB according
to an
article
on dealnews.com.
January 2013 -
Micron announced it will
ship SATA SSDs with just undet a terabyte capacity priced at under $600 in
this quarter.
April 2013 -
Astute Networks
said its new fast-enough 2U iSCSI
SSD array can deliver virtualized deduped storage capacity at $2,000 /
TB.
March
2015 - SanDisk
said its InfiniFlash (a 3U 512TB rackmount SSD) costs less than $2,000 / TB -
before compression or dedupe.
January 2016 -
Mushkin said it will
ship a fast-enough consumer SSD with 4TB capacity costing $500 in mid 2016
according to
TechReport.com | | |
| . |
|
|
| . |
| Many SSD marketers I
talk to nowadays regard the price per gigabyte of their latest product almost as
a badge of honor. But it wasn't always like that. SSD pricing used to be a taboo
subject. |
| Breaking Taboos on SSD
pricing | | |
| . |
| Cactus looks at the thorny
issue of embedded flash TCO |
Editor:- April 2, 2014 - Cactus Technologies
today published a blog -
Solid
State Storage Total Cost of Ownership versus a Really Low Price Today -
aimed at designers in industrial markets - which discusses 4 sources of cost
they should consider when selecting an SSD.
When looking at eol
considerations - the author Steve Larrivee
- warns that although designers may be counting on being able to delay
requalifications by mining obsolete SSDs as unsold inventory from channels and
brokers "for a considerably higher price... this introduces the
possibility of counterfeit parts as well."...read
the article | | |
| . |
$/TBW (cost per Terabyte
Write) was proposed in a blog by SanDisk.
There are many other
misleading metrics like this in the SSD cost evaluation literature. And you'd be
surprised how many of them originate from leading SSD companies. |
| cost per Terabyte Write? | | |
| . |
|
| . |
|
|
. |
|
|
| . |
|
|
| . |
| IOPS / $ as a value of
SSD goodness is bad |
Editor:- December 5, 2012 - The
cost of SSDs is one
of the arguments most often cited by antis to explain why (in their view)
the transition to a pure SSD storage market can't happen.
I guess
the designers of the first ships made from iron (which unlike wood doesn't
float) and the first airplanes (which were heavier than air) must've got used to
hearing similar objections.
In the past 10 years in various articles
I've described what I thought were the real
user value
propositions for SSD adoption in different markets. And I've got so used to
filtering out the inappropriate arguments about the cost per gigabyte of SSDs
which come from the SSD antis that they don't bother me any more.
But
the broad sweep of principles which govern SSD adoption in the enterprise still
leave much room for misleading analysis - and even when the analysis comes
from SSD vendors - then it's just as important in my view to show that it's
wrong.
So be wary of arguments for enterprise SSD adoption which cite
IOPS per dollar (or the other way round) as a justification for filling a
gap in some cleverly drawn curve. I've seen this recently from leading SSD
companies who should know better.
As human beings we feel comforted
when we think we see new patterns. But they don't always reflect reality.
SSDs fitting gaps in IOPS vs dollars charts isn't a sufficiently good
reason to buy SSDs. Just as buying stocks based on past performance charts
isn't a good idea either. The future isn't a tidied up remake of the past.
If
you think about it for more than a microsecond you'll see that the IOPS per
dollar argument - which sounds plausibly eco-technical when you first
see it written down - doesn't lead to any safe conclusions at all.
the
zero cost but slow SSD fails this test
Apply some
boundary
analysis to the situation and imagine an SSD which costs zero dollars.
According to the IOPS/$ advocates - this is your perfect enterprise solution
and at this cost it should even replace hard drives. But if the IOPS of the
SSD is considerably less than that of the hard drive - you'd be nuts to use
it - because you wouldn't get useful work done on your apps. (The model has
broken down in this direction.)
ultrafast SSDs with low capacity
fail this test too
At the other end of the scale - it's easy to
picture ultrafast SSDs which could score very well on the IOPS / $ scale - but
whose capacity or electrical power consumption or or physical size or
reliability doesn't make them scalable or attractive solutions.
I said
recently on these pages that if you're going to try and come up with a single,
plausible-sounding eco-technical marketing-jargon metric for SSD adoption
in virtual server environments - then a better suggestion is "cost per
happy virtual user". That's the total system cost BTW (TCO) - which
includes all the servers and software licenses and service costs etc etc.
IOPS
per dollar is a useless metric
Once you start looking at complex
real-world systems - which run more than a single type of app - you'll see that
the only way in which users can optimize their total economics is by
mixing and matching different speeds of SSDs in what I call different
silos in their apps
server architecture.
Even if you haven't read that read that
article it's easy to grasp why I say you can only get the truly lowest
system cost by using different types of SSD with different speeds,
different capacities and different operating characteristics (and certainly
different $/IOPS) - within ratios which make sense for that type of app mix and
user load.
$/IOPS is an illusion which doesn't take you anywhere far.
Now
- everything above - up to this point here - is something I've already said or
implied before - so what can I leave you with on the SSD news page today which
is new?
Well here's something I've been thinking about.
Faster
SSDs may reduce the amount of SSD capacity you need in a well designed VM
environment - compared to using slower SSDs - for the same number of users. The
reason is that while there is some overhead in capacity which you can't avoid -
on a per user basis - such as their unique data - there's also a lot of
transient system storage capacity which gets allocated and deallocated
dynamically to get their work done.
As we know in the hard drive VM
world - at the threshold of usable performance you need more capacity (really
- more drives) simply to get enough IOPS performance. The HDD experience is
the best sales person for introducing SSDs.
But when you're comparing
different types of SSDs in a VM situation - then the software makes a big
difference too. If an SSD from vendor X is twice as fast as that from vendor Y
- it means that SSD-X frees up some of its capacity faster than SSD-Y and can
therefore handle more users for the same purchased capacity within the same
elapsed time.
How big is that percentage difference?
I'm sure
we'll see more competing claims coming out from the
SSD software industry
in the months ahead. | | |
| . |
|
|
| . |
|
|
| . |
|
|
| . |
|
|
| . |
| |
