7 ways to classify where all SSDs
will fit in the pure SSD datacenter |
by Zsolt Kerekes,
editor - March 27, 2012
I get many questions asking me to
explain how the various pieces of my enterprise SSD market model and
architecture models fit together to help them understand some basic questions
like this...
- what are the main types of SSDs that we'll see in the future enterprise SSD
market?
- which companies are the likely leaders in these future segments?
- is SSD company X - a leader or a future drop out?
- how much is this company / technology / market segment - really worth? -
and for how long.
- do these companies X and Y really compete with each other? - if one wins -
does that mean all the others lose?
- if I bought this SSD company - would it be worthwhile?
You can see
how this goes...
The purpose of this new article is to provide
a top level summary of the smallest number of usefully different types of SSD
system which can coexist sustainably and inter-operate in distinctly different
roles in the architecture of the pure solid state enterprise datacenter of the
future.
Most of the interface types and form factors and memory types
for SSDs have been around in some form for many years. This isn't about
morphology. Although shape and interface type do come into the scope of the
classifications below - when they are relevant.
SSD application
silos within enterprise architecture
When thinking about where SSD
products fit into the pure solid state storage enterprise datacenter
architecture it can be useful to think about them from the point of view of
application silos - and where they sit relative to the application processors.
Vendors
- who by design (or accident) - offer products which match the specification and
price contours of these SSD silos - will have advantages over competitors whose
products don't.
SSDs aren't islands - their data always comes
from and goes on to other SSDs
Enterprise SSDs which have been
designed without any reference to how they will segment and interact with other
types of SSDs - in the mind's eye of the customer architecture - will
eventually fail in the market - because they will have the wrong set of
technical features and therefore cost more to satisfy the genuine SSD
architecture needs of the market.
An example here would be a "fast
SSD rack" designed for apps acceleration - but which also included dedupe
and compression. These are inappropriate features for this class of product.
Such a product set may work for some customers - but mismatched feature sets are
not scalable over time and over bigger markets.
The enterprise SSD
market is complicated enough already and users will steer towards a simple set
of standard product types which meet universal needs and which can be almost
seamlessly replaced by similar products from competing vendors - without
re-engineering the user's legacy SSD architecture.
All server SSDs can
be described as fitting into one or more of the following categories.
- PCIe SSD cards /
modules (or whatever replaces PCIe in the future as a generally
available lowest latency, highest bandwidth connection between SSD
memory arrays and server CPUs - such as memory channel SSDs
).
App silo = acceleration SSD (internal).
Lives closest to
the application server (and often inside the server box)
Why do people
buy them? Because they provide easy to understand, easy to scale and low
cost ways to increase virtual server apps performance - by placing useful
amounts of low latency storage transparently close to the CPU.
Not all PCIe SSDs are
the same. The PCIe SSD market will, itself fragment into about many
different types. ultrafast, fast, fast enough, ASAP, legacy vs new dynasty, and
standard reliability vs high availability. And the PCIe attachment concept is
also stretching into the popular removable
2.5" form factor
too.
Examples of companies who offer products in this form factor
include:- Fusion-io,
Texas Memory Systems,
Virident Systems,
OCZ and
LSI. About 30 other
companies are listed in the
PCIe SSDs directory.
- ultrafast rackmount SSDs. In today's market the
fastest
rackmount SSDs on
the SAN are always RAM SSDs. So I was tempted to just label this category "RAM SSDs" - but that
would be misleading because RAM SSDs are available in other form factors too -
such as 3.5" and
PCIe cards - so that wouldn't be so clear.
Whatever the memory
technology - the ultrafast rackmount SSDs provide ultimate low latency and
the highest random IOPS for any size of blocks.
App silo =
acceleration SSD (on the SAN).
Why do people buy them? (Given that they
are the highest cost storage per terabyte and lowest density per rack unit)...
RAM
SSDs solve some bottleneck problems which are either impossible to solve with
other SSD technologies - or which would cost much more by requiring more servers
combined with over-provisioning of fast flash SSDs for example.
Examples
of companies in this category include:-
Kove,
Kaminario and
Texas Memory Systems.
- fast rackmount SSDs. In today's market these are the fastest
flash SSD systems which you see from companies like
Texas Memory Systems
and
Violin. Their primary
characteristic is speed. And they connect via traditional storage networks
like FC SAN ,
iSCSI,
IB and even
PCIe.
App
silo = acceleration SSD (on the SAN).
Why do people buy them? - To get
very high performance. But most enterprises don't need this level of performance
and even enterpirises which can benefit from these fast systems can't afford to
implement all their SSD storage exclusively with these fast SSDs. That's why
there's another distinct type below.
- fast enough rackmount SSDs. These are SSD arrays which are 5x to
10x slower than the fastest "fast SSD racks" in the same memory
technology.
App silo here depends on the enterprise. It can be viewed
either as the main form of acceleration SSD on the SAN (in situations which
don't need the faster SSDs) or it can be viewed as a distinctly different
SSD performance layer on the SAN (auxilary acceleration SAN SSD).
So
why would anyone buy them? Because they're significantly cheaper - for the same
capacity. They're cheaper - because they are much easier to design than the
fast systems.
Fast-enough SSD racks still provide useful speedups
compared to legacy HDD arrays, and are faster than bulk storage SSDs (below).
For many types of apps - this is as fast as the rackmount SSD has to go. Faster
operation would be economic overkill for the business. Like all rackmount SSDs -
they can co-exist in the same enterprise with PCIe SSDs (which are essentially
part of the server).
Examples of companies in this category include
Texas Memory Systems
and
Violin who have
diversified their product lines to offer cheaper / slower models. It also
includes companies like WhipTail
Technologies, Nimbus
Data Systems and Pure
Storage.
- capacity
maximized SSD racks / bulk storage SSDs / archive SSDs /
cloud storage SSDs.
These
systems provide the maximum amount of SSD capacity per rack unit at the lowest
operating cost (and lowest electrical power). But their latency and random
IOPS is orders of magnitude slower than fast SSD rack systems.
So why
would anyone buy them? Because they provide the lowest cost usable storage on
the planet (even lower cost than HDD arrays). The main applications are
backup and archival
storage.
A typical feature of this type of SSD system is transparent
integrated compression and deduplication. In other types of SSD rack that would
compromize performance - but in these systems capacity density has the highest
design rating.
Vendors in this category include:-
GreenBytes and
Skyera.
- SSD ASAPs (auto
accelerating / caching / tiering SSDs). The main role of these storage
appliances is to intermediate between the top levels of installed SSD speed
in the enterprise.
Why would anyone buy them? - No single SSD type can
match all the needs of all user enterprises economically. And there will always
be a need to have intermediate management between SSD systems which have
dissimilar speed / cost characteristics.
- high
availability / fault tolerant SSDs. Unlike standard SSD systems - which
may have some fault tolerant aspects in their design such as
RAID - the HA SSDs include
significant features in their designs which support performance while offering
no single point of failure. Performance - and in particular latency - is not
significantly degraded in the event of sub systems failure. That's in contrast
to other methods which wrap failover support outside the SSD box - and can
destroy the performance advantages of the SSD when it's in safely failed mode.
HA
SSDs can be classified as fast or fast enough.
App silo = acceleration
SSD (on the SAN).
In the current market there's no point in vendors
offering HA SSDs with lower performance - such as for archiving. Bulk storage
SSDs will have replication features already built-in - as it's relatively cheap
to achieve this using traditional methods which are already compatible with the
slow latencies offered by such systems.
This is where the 7 distinct enterprise SSD silos list ends. It's
only if you're being really picky about completeness - that you might also want
to find a place for 2 other categories below - which aren't really part of the
SSD app silo model.
- component SSDs. This isn't a new classification. But obviously
most of the above rackmount SSD types for example can be implemented
internally by arrays of smaller SSD modules such as 1.8" SSDs, 2.5"
SSDs and even PCIe SSDs. These SSD components can be used inside what I call
"open" or COTS (commercial off the shelf) SSD arrays - which can be
engineered to meet any of the above purposes.
From the marketing point
of view there are already many long established and useful ways to segment the
SSD components sub-markets - such as form factor, interface, data integrity,
security, industrial, military, notebook, enterprise etc - such as in our SSD
buyers guides.
There's an overlap between SSD vendors who market SSD
systems and those which only market components for so called oem use - (use by other
manufacturers). From the user's point of view these differences of
sourcing are most apparent in the "fast SSD rack" category - and least
visible in the "fast enough SSD rack category". These differences to
do with SSD controller
design and big
versus small SSD architecture - have been dicussed in other articles -
so I won't repeat them here.
- notebook
SSDs - It doesn't add anything to our understanding to include notebook
SSDs within this classification article.
Do PCs, notebooks, netbooks,
tablets, phones etc form part of the enterprise SSD population?
Yes
of course they do - because most of the demand seen in the application servers
originates from people using (for want of a better word) PCs. But from the
marketing point of view the segmentation of SSDs within these devices is best
dealt with at the component level. And there are plenty of
SSD analysts who
can tell you anything you can afford to know about the SSD notebook market.
Enterprise
buyers don't choose their notebooks because of how well their internal SSDs
interact with their server SSDs. The choice of notebook SSD is irrelevant to the
datacenter architect. Instead it's the pattern of the total data demand
originated from all the notebooks which is the key factor. There's a disconnect
when we get to this level. And that seems like a good place to finally
end these notes.
For more related enterprise SSD articles - take a look
at the list I picked in the right hand side of this page - or for other
recent SSD blogs see the index page
(home page) of StorageSearch.com or take a look at the
most popular SSD
articles seen by readers in the past month.
See also:-
Can you
trust SSD market data? where are we now
with SSD software? Recent Strategic
Transitions in the SSD market understanding flash SSD
performance characteristics and limitations | |
. |

| |
.... |
 |
|
Is there room for me too? |
| |
.... |
 |
. |
Decloaking hidden segments
in the enterprise |
Some of the world's leading SSD
marketers have confided in me they know from their own customer
anecdotes that there are many segments for enterprise flash arrays which
aren't listed or even hinted at in standard models of the enterprise
market.
Many of these missing market segments don't even have
names.
Hey - that means SSD-world is like a map of the US before
Lewis and Clark.
If you're a
VC should this make
you anxious or happy?
If you're a user - maybe that's why no one is
delighting you in the way you think you deserve.
That's what led me to
write my new article -
Decloaking
hidden segments in the enterprise for rackmount SSDs | | |
. |
DIMM wars are the new
territory of focus for the development of the fastest server based SSDs -
following a 7 year reign of supremacy by the PCIe SSD market (from 2007 to
2014).
What we've been seeing in the enterprise PCIe SSD market in
recent years is that most vendors are working on ways to make these products
more affordable rather than trying to push the limits of performance. |
DIMM wars in SSD
servers - Memory1 - episode 1 | | |
. |
 |
. |
|
. |
SSDserver rank is a latency
based configuration metric - proposed by StorageSearch.com - which can tersely
classify any enterprise server - as seen from an SSD software perspective - by a
single lean number rating from 0 to 7. |
what's in a number?
| | |
. |
 |
. |
|
. |
|
. |
DWPD (Diskful Writes Per
Day) for 5 years - has become an established part of SSD jargon in the writings
of enterprise SSD makers in recent years. |
what's the state of DWPD? | | |
. |
let's hear a cheer
for
the "fast-enough" enterprise SSD |
It's difficult to
overstate the importance of the "fast-enough" enterprise SSD as a
new emerging market segment in the market in the 2012 to 2016 period.
How
do you recognize one?
At its simplest - the fast-enough PCIe SSD today
- for example - offers similar performance specs to the fastest PCIe SSD from
4 to 5 years ago.
Some differences are that today's fast-enough new
model occupies less slot space and costs much less than the old fast model
- due to advances in memory technology. But for many users, and in many
applications - the fast-enough SSD product is all they need (at the right
price).
In
earlier
phases of the SSD market - the
prime justification
for buying SSDs was speed. In those days - if an enterprise SSD vendor didn't
work hard enough to compete with the
fastest - and
merely offered the type of speed which years later would be acceptable in the
fast-enough category - they became toast - because no one was going to buy an
expensive acceleration SSD with less than (contemporary) best in class speed
specs.
In the SSD market of today and in the future the affordability
and utility of SSDs at the lower end of the performance spectrum creates a
distinct new market - because SSD buyers have a choice - and the cost of
acquiring new SSD customers is much lower today than in the past - due to the
fact that they are already tuned into this market and seeking SSD solutions.
Lower cost enterprise SSDs weren't a viable business model for
vendors in the past due to the much higher cost of
educating and
acquiring new customers.
(Viable SSD business models aren't just about the cost of
memory chips and
controllers.) | | |
. |
 |
. |
| |