| leading the way to the
new storage frontier||...|
top SSD companies
what's the state of DWPD?
popular SSD articles
hold up times in 2.5" military SSDs
What were the big
SSD ideas to learn and forget in 2015?
is remanence in
persistent memory (NVDIMMs) a new risk factor?
|is it realistic to talk
about memory IOPS?|
|Editor:- August 11, 2016 -
million IOPS on a single device is the title of a recent blog from Storage Switzerland which is a
briefing note about Crossbar
which operates in the conerging segments of the
alt nvm and
Among other things author - George Crump - says "Crossbar
believes that it can achieve 24 million IOPS on a single 4TB NV-DIMM without the
use of a RAM buffer or a capacitor." ...read
Editor's comments:- when startups enter new
emerging markets they are often tempted to make headline grabbing claims.
And I think the "24 million IOPS" (IOPs as in you
and I think about them) has to be interpreted in that context. (How can you
claim record breaking IOPS when all you've got is a memory IP - and that's just
part of a yet to be integrated technology set which together make IOPs.)
is not to decry the importance and validity of the
change in the SCM SSD DIMM wars market - which have consumed nearly
half of my working hours in the past year.
We saw similar wild
claims when the startup Fusion-io
was trying to get across how
PCIe SSDs would change
the enterprise storage market by reference to the nearest similar technology
when Fusion said in 2007 it would replace SANs. (Because SAN based SSD
accelerators were at that time the SSD market's dueling weapons of choice.)
back to Crossbar - there is a genuine problem for the industry (which I touched
on in an earlier post about Diablo's
DMX software) - which is - what are the most useful metrics to judge tiered
memory systems by?
As we've seen in the
SSD accelerated storage
pool market since 2009 - there's a wide spectrum of use cases and cost
considerations which have many viable business intersections.
need new "goodness" numbers for DIMM wars memories.
think using IOPS to characterize a memory product is less useful to describe why
people might want to look at it than wattage, raw capacity in a DIMM,
uncached raw R/W latency and price.
And - most important of all -
what software does it work with? And how well does the software behave?
Development Board - pics on Tom's Hardware|
|Editor:- August 12, 2016 - A year after Intel and Micron preannounced
their 3DXPoint / Optane bid for the
market (with the absence of any worthwhile technical details) we now know
that you might have to wait another year to buy any.|
But the companies
have released more preliminary tech data and this week have been showing
prototype development boards at FMS.
can see a useful article about this -
XPoint SSD Pictured - on Tom's
Among other things this says the new memories will
have an endurance rating of 25 DWPD.
speed will not be fast as (selectively) promised last year.
shouldn't come as any surprise to regular readers.
June 2016 I
recommended a video 3D
XPoint, reality, opportunity, competition by Yung Lee,
President & CEO - BeSang - who
correctly (as it turns out) reinterpreted the performance claims of 3D XPoint
and looked at the limitations, manufacturing costs and best likely application
roles for this new technology from a deep semiconductor perspective.
is Micron rebranding the memory to QuantX? - the 3rd name for the same memory
Well why did Microsoft skip ahead to Windows 10?
Branding SSD related
products is not as easy as you might think.
But it could be that by
the time anyone can see these new products in an online shopping basket the
original name will seem old.
persistent memory pose new new security risks?|
|Editor:- August 4, 2016 - 35 years ago I noticed
something curious about RAM which surprised me. But it wasn't important for my
work at the time. |
In a new blog on StorageSearch.com -
remanence in persistent memory a new risk factor? - I link together some
ideas which pinged across my path 3 times in as many decades. But I think these
have more relevance now in the new context of emerging persistent memory
|fast is good - more is
|Editor:- August 9, 2016 - In some ways Mangstor has some
quaint old fashioned ideas because its products are to be found in the
segment of raw enterprise performance and you'll find them listed along with
some others in my indicative article -
the fastest SSDs.
Speed was indeed the gateway which opened the access to the
enterprise SSD market back in the days of
although just as with cars - having long ago proved that
SSDs can run
faster than hard driven horses - the enterprise market has recently grown
more sophisticated in its analysis of
requirements - even to the extent that vendors are now offering pricing
schemes which are
far away from traditional cost for capacity ideas of rotating storage.
brings me back to Mangstor - a company I described last year as having "less
employees than cores in its SSD controllers."
There is another
old fashioned business idea which we've seen has to be satisfied for any such
company to succeed for long in the market. They have to be able to answer this
question - posed by their ideal hypothetical customers...
like what your product does - but do you have an easy way we can use a lot of
them together - which preserves the essential character of
like - without crossing over into product territories which other vendors do
Many products which look like stars as solo SSDs fail such
Mangstor's answer to such a question can be seen in an
today - " the availability of its TITAN NVMe over Fabric (NVMf) Target
Software Storage Stack which tightly integrates NVMe SSDs with Remote Direct
Memory Access (RDMA) NICs and efficient use of x86 server capabilities to
deliver network storage performance with upto 10x higher bandwidth and
1/10th the latency compared to iSCSI and FC-based AFAs."
Mangstor also says - its NVMf arrays based on TITAN software scale to
petabytes of capacity and hundreds of GB/s as arrays are added with virtually no
increase in latency.
In case you didn't see this article earlier - the
value of low latency in enterprise SSD contexts was elegantly summarized in
the SSD Bookmarks by
|lowering cloud wattage
with low DWPD SATA SSDs|
|Editor:- August 4, 2016 - Although it's the
faster SSD products (like PCIe
SSDs and memory
channel SSDs) which capture the attention of readers - because they show
what is possible (and after a long enough interval we see pioneering
enterprise speeds becoming commonplace at lower
prices as we're now
seeing with M.2 SSDs)
nevertheless - when it comes to where most of the SSD slots are - the workhorse
of the SSD market - in arrays, webscale and cloud - is still the simple
Well, maybe not that simple - because since about
started to see subtle power optimized and mostly read oriented (low
DWPD) SATA SSD product
lines being introduced specifically for use in dense populations in the
big market for SSD
vendors and SATA SSDs
are a low risk choice for users because there are so many competing companies
and products that ensure continuous improvements in value and quality.
new addition to this crowded market is the -
XF1230 (pdf) - a 1.9TB capacity SSD which consumes less than 5W, is
rated at 0.67 DWPD -
will ship to channel partners next month.
|FIPS makes for health
|Editor:- July 28, 2016 - A new
by Micron -
discusses FIPS 140-2 validation in its
of SAS SSDs.|
The author - Anne Haggar, Product
Marketing Leader, Micron says (among other things)...
finding that U.S. federal agencies arent the only organizations that are
interested in the extra security these drives provide. Companies in health care
and financial services who face stiff fines for non-compliance and huge risks if
they have a data breach are adding FIPS 140-2 compliance to their requirements."
|Non-Balanced Wear Leveling
- a paper by Renice|
|Editor:- July 28, 2016 - Renice Technology
recently published a paper -
Wear Leveling Algorithm (pdf) - which outlines the thinking behind a
specific technique in its industrial SATA3 SSD controller - model
- to improve
upto 3x compared to traditional methods. |
This is one of the
several techniques used in this controller to overall get a 20x
improvement in lifespan when using MLC. ...read
the article (pdf)
Editor's comments:- Ever since the
first flash devices were evaluated it has been known that some blocks are much
better than others.
As an example in this paper Renice shows that
in a modern 16GB MLC flash chip - even after just 10 P/E cycles the controller
is able to see a 3x difference between the fastest and average program time and
over 30% difference between the slowest and fastest read times.
The quality of wear resistance tells you something which
can be used to grade blocks.
Renice's non-balanced wear leveling
algorithm leverages these naturally occurring process variations so that
"the higher wear resistance blocks are selected to be erased more times
while the lower ones get protected instead."
Although there are
no fundamentally new ideas presented in this paper - because the technique is
just one permutation of many from the superset of all
techniques - this paper does provide a useful survey of classical wear leveling
techniques along with their associated trade offs in performance and endurance.
I got a good sense of judgment and balance in this paper - given the
Context is always important - and these techniques
are discussed in the context of general purpose, simple low power
which use modest speed
That's distinct thinking from new generation enterprise
array controllers in which visibility into other SSDs in the same array,
larger ratios of DRAM and knowledge about the applications software stack can
also be leveraged to reduce endurance.
Here are more articles on
|data dematerialization in
|Editor:- July 27, 2016 - Some of the big SSD
ideas in recent years have been:-
One way to interpret the essence of Symbolic IO's architecture - which was
partially unveiled in May
2016 - may be as a coming together of the 2 concepts in the same place...|
got me thinking this way was a recent blog -
look at Symbolic IO's patents - by Robin Harris on
his site - StorageMojo.com .
Symbolic IOs founder Brian Ignomirello who saw
and liked Robin's post - said among other things on
- "yes we (do) materialize and dematerialize data." ...read
PS - In a conversation I had about the market
yesterday (which I'll write about next month) I noted how during the past year
the SSD industry has been thinking much more about
systems architecture as the next emerging core for innovation in the same
kind of way that the market was buzzing about apps acceleration using SSDs on
the SAN and in servers via PCIe cards 8 to 10 years ago.
of all that previous market experience with
PCIe SSDs in servers
especially and the comparisons with other ways of getting similar results with
arrays of SAS / SATA SSDs in storage - and the feel-good confidence from having
made those difficult changes - that the data computing market is now
receptive to being more ambitious with re-engineering memory.
business incentive being that the gap between what is possible and what is
being done every day with current products is so huge and
(Huge savings for users. Multi-billion dollar new markets for vendors.)
already seen a lot of different approaches coming down the pipe in the past year
with technology announcements. But even though the implementation details are so
different - they're tackling the same problem.
With more toys in the
memory, SSD and software tool kit - there are now more permutations for
servers which exceed previous performance limits and slash away at legacy
The DNA of the semicondata market continues its
ruthless quest of doing more for less.
about RAM being able to store only a limited amount of data are becoming a
non-issue. Recently, Amazon introduced their X1 Instance, which lets customers
rent computing instances with 2 terabytes of RAM for just $13.34 per hour. At
such pricing levels, the economics of renting large clusters containing
virtually any amount of RAM are well within reach."|
President & CEO at GridGain Systems
in his blog -
Computing Will Dramatically Change Our Lives (July 14, 2016)|
|new industrial secure eUSB 3.0 10-pin SSDs
August 22, 2016 - Virtium
new models of eUSB 3.0 10-pin models in its TuffDrive SSD range which offers
much faster speed than eUSB 2.0 at about half the power consumption.
range from 2GB to 256GB, while drawing less than 1W. Size is 36.9mm x 26.6 mm.
also:- USB SSDs,
Plexistor demonstrates low latency memory fabric on commodity
Editor:- August 19, 2016 - If you don't want to be tied
to any particular proprietary big memory hardware - but do need memory fabric -
what kind of performance can you expect using
recently demonstrated it can handle millions of remote writes per second at
latencies as low as a few microseconds in a benchmark on Mellanox
infrastructure over 100GbE using Plexistor's persistent memory over fabric
Benchmarks performed on a Mellanox infrastructure over 100GbE
using Plexistor's PMoF Brick demonstrated record performance: more than 1.6
million random 4KB IOPS at less than 6µs with throughput of 7GB/sec.
where memory goes - compression follows
August 19, 2016 - Cache memory compression techniques and a survey of why and
where they can be useful featured in a
news story here
in April 2015
and was part of the rethinking RAM
in an SSD context industry trend which is sweeping across computer
But where can we find examples of such techniques being
A recent case is a
of software from A3Cube
- which relates to their RAM over FABRIC system. Among other things -
Fortissimo Foundation 2.6 now includes In-Memory compressed IO caching.
we're #1 in enterprise SSDs - says Samsung
August 11, 2016 - which SSD company is the market share leader in enterprise
SSDs? (Drives not racks.)
claimed the #1 crown in a recent
which quotes various percentages (32% to 45%) from 3
market research companies
which relate to the first quarter of 2016.
In part Samsung may be reaping some benefits from the frenemy trend in the
enterprise storage systems market.
This is the change in supplier
relationships when SSD drive companies begin to sell complete rackmount SSD
systems as part of their product portfolios. While Samsung remains a component
company and not a systems company (and the risk assessment of its SSD oem
customers suggests that Samsung isn't a threat to their own business) this is
a good enough reason for many array vendors to prefer Samsung as a drive
Among other things in my
2015 big SSD
ideas roundup article I said this...
which once appeared safe...(list) are no longer safe assumptions when your
biggest supplier can become your biggest competitor. Or when your biggest
customer can become your competitor."
In detail the frenemy
concept is more complex - because when you look at the range of
in the SSD enterprise and the differences in
latency tiered products
and IPs needed to become a leader in each of those segments - an experience
supplier risk assessment can be finely grained - as in "this company has
systems products for webscale and cloud - but my market is traditional mid range
enterprise - which requires more services and software - so therefore such a
supplier is not an immediate threat to my business."
assessments have a short life in today's market however as everything will
change due to growing user sophistication and dissatisfaction - as I discussed
in my article -
of the enterprise SSD companies which you know have no good reasons to survive.
But for now - and not withstanding the
of inferring too much from SSD market data - and given the new trend for
tiered memory systems to change the
enterprise SSDs in server and storage systems - any company which has a leading
market share position in enterprise SSDs today has to be congratulated with
the warning that everything is still changing and the rate of
PS - the ability of past SSD shipment data to predict
long term future success is something which I called into question over 9 years
ago when I started publishing the
Top SSD Companies series.
- dear readers - still have the ability to change markets, surprise us all and
remind us that the SSD architecture idea stands out as
1 of the 3 main
disruptive changes in the computer business in the past 40 years.
SMART samples 32GB DDR4 hybrid DIMMs
9, 2016 - SMART Modular
imminent sampling of its 32GB DDR4 NVDIMM-N (hybrid DIMMs).
comments:- Having recently written my blog -
remanence in persistent memory a new risk factor? - I asked SMART - "Does
your IP offer the option to encrypt data backed up to the flash in the NVDIMM?"
who is Director Product Marketing at SMART told me - "Yes, it does have
the capability to do encryption, but it has not been enabled. We are in
discussions at the JEDEC
meetings as to how this is implemented with the NVDIMM specifications."
Nimbus re-emerges from stealth with 1PB / U raw HA SSD
August 9, 2016 - Nimbus
Data Systems has emerged from its self imposed exit into
mode with the
of a new range of Ethernet/FC/Infiniband attached rackmount SSDs based
on its new
OS with GA in Q4 2016.
Entry level products start in a 2U box with
50TB raw capacity for under $50K and for larger configurations Nimbus says its
ExaFlash offers an effective price point as low as
effective gigabyte (including all software and hardware).
density boxes in this product line - D-series models - will have 4.5 PB raw
capacity in 4U (12 PB effective).
- I haven't seen details - Nimbus says there is no data network between the
storage arrays themselves, guaranteeing that performance truly scales in
lock-step with capacity and with consistent latency.
video above shows the 4PB raw ExaFlash
Editor's comments:- if there are to be sustainable
roles in the future
enterprise SSD systems market for AFA vendors which previously sold arrays
of SAS/SATA SSDs - and who don't own their own semiconductor fabs - the only
viable ways to establish such platform brand identities are with SSD
software and architecture.
There's a huge gap between the technological
aspiration which Nimbus talks about and the weakness of its past marketing and
the kind of
funding which we've
seen competitors in this market burn through in the past with mixed results.
In the next few quarters I hope we'll hear more from Nimbus about its
business development plans and customer adoption.
roadmap to the
positioning of many flash array "startups"
Liqid controller inside fastest 2.5" NVMe flash SSD
August 9, 2016 - In a joint
release today Liqid
gave details of the "fastest 2.5" PCIe NVMe flash SSD ever
benchmarked" - which they're showing this week at FMS.
2.5" PCIe SSDs,
- upto 3.9 TB of capacity
- Mixed R/W: ~ 5.5 GB/s (Full Duplex)
- Seq read ~ 3.6 GB/s Seq write ~ 3.6 GB/s
- mixed random R/W ~ 1.15 M
IOPS (4 K,
Seagate previews 60TB 3.5" SAS SSD
August 9, 2016 - Seagate
it 's demonstrating prototypes of a 60TB 3.5" SAS SSD which will be
available next year.
new memories? new security risks?
Editor:- August 4,
2016 - Is remanence a security risk in persistent memory? That's the topic of
my new blog
you aren't yet ready to evaluate these new SCM style NVDIMMs you might think
you can skip this article.
That's OK as long as you already were aware
that that data recovery
has always been feasible in old style
DRAM too. ...read the
former SanDisk CTO joins 3D fabless BeSang
August 2, 2016 - today it was
that Kevin Conley,
former CTO at SanDisk
has joined BeSang as a member of
advisory board where he will help guide the company to find SSD applications for
their 3D Super NAND nd 3D
Super NOR technologies - which unlike conventional through-silicon
vias 3D can "stack high-density, multi-memory layers sequentially on top
of other device layers in a single chip at low cost using proven material and
Radian will sample PCIe based competitor to NVDIMMs
August 2, 2016 - Radian
it will be sampling in October a new byte addessable,
(NVRAM-flash cache ratio), 12TB
PCIe SSD which has
on-board host controlled tiering between its flash and NVRAM.
comments:- Radian is positioning the new product as a cost effective
alternative competitor to
hybrid NVDIMMs and
similar emerging products in the
SCM DIMM wars
In an interview last year with Radian's CEO Mike Jadon -
what's the role for a
Radian Memory SSD? - I learned more about the company's software thinking.
Diablo gets more funding for Memory1 and DMX software
August 2, 2016 - Diablo
it has secured $37 million across 2 phases of an oversubscribed round of Series
New investors Genesis Capital and GII Tech Ventures
joined the second phase of the round, along with follow-on investments from
Battery Ventures, BDC Capital, Celtic House Venture Partners, Hasso Plattner
Ventures and ICV.
Editor's comments:- that's kind of
interesting - but much more interesting from my perspective was what I learned
in a 1 hour conversation with the company last week about the software for their
Memory1 (flash as RAM) product.
Diablo's DMX software is barely
mentioned in the funding PR above. I expected to see more on their web site
I also learned how Diablo handles the flash and endurance
Those aspects were mysterious to me when the product was
announced last year. But it's very straightforward. I'll write about them in an
article later this week.
Until then - if you're wondering - the best
way to think about the caching and tiering side of things is that Diablo's
software leverages DRAM on the motherboard.
This DRAM (in another socket) must be present for every 1 or 2
Memory1 modules in the system. And in many respects it uses that DRAM and its
own flash in a similar way
to the early Fusion-io
PCIe SSDs and some of the other tiering, caching products we've seen before like
Diablo's DMX operates in memory layers and also the company has done machine
learning of popular and proprietary apps it might work with so that it
understands the nature of data demand patterns and structures.
says that - unlike NVMe and those other
storage cache / tiering
products - the benchmarks they've done with Memory1 have much more
acceptable operation - because DMX and Memory1 don't have the same variability
of latency which occurs when you go through storage stacks and storage or
This inconsistency of latency is one of the
problems I wrote about in my classic article on
symmetries. And the consistency of - for example -
- was a powerful competitive difference exploited by marketers at enterprise
PCIe SSD pioneer - Virident.
There are always infrequent traffic related congestion and contention
problems in any multi-tiered latency system - even in real world physical
These rare clogging events (nanoseconds,
microseconds, or milliseconds) accumulate to bigger actual latency numbers in
storage interchanges. So when you're emulating memory with storage - it's not
the best performance which matters. It's the worst case which causes QoS
Diablo says the ability of DMX to understand data at the
application level and move it between DRAM and flash via the DRAM bus with
native custom silicon
controller support for
these memory movements - gets results which are on average several times better
than the best average PCIe based flash cache alternatives. As you'd expect.
But it's the superiority of the worst case latencies - which can prove
to be the yea or nay breaking point in the selction between DIMM based
flash and other interfaces for critical memory emulation deployments.
includes a QoS latency feature so that application developers can retain
control of data they like in DRAM without having to rely on caching
More from me on this and the endurance side of things
relating NVMdurance's machine learning to manual tuning
July 29, 2016 - Nearly every SSD in the market today from the smallest SSD on
a chip to the bewildering array of rackmnount systems can be viewed as a choice
of how to select and mix the raw ingredients of SSD IP and integrate them into
products which (for better or worse) match up to and satisfy user needs. How
these decisions are made depends on the DNA of the product marketers, the
technology teams, familiarity and ease of access to some technologies rather
than others, business pressures and timing, the willingness to take risks, and
sometimes - just luck.
But all products - no matter how complex they
appear - can be analyzed as a specific set of choices made from the architecture
and IP selections which are possible.
In many articles in the past I've
shown you how - whether you're looking at the design of SSDs or systems - there
are rarely more than 2, 3 or 6 raw available decisions which determine each
piece of the jigsaw. And I know from the feedback I get from SSD specifiers and
architects that these simple classifications can be useful in helping to compare
different products and even in choosing which competitive approaches are similar
enough to make comparisons worthwhile.
But when you get down into the
details of implementation at each layer in the product design - every one of
these dimensional options which go into the permutations blender to shape the
total product identity - can itself be complex and multilayered.
the example of the raw magic tuning numbers which enable the raw R/W program,
erase, threshold voltages, shaping and timing parameters inside a flash memory.
The question of how much and when has been at the heart of what makes some SSDs
better than others ever since flash was first used in SSDs.
designers have spent their whole careers measuring and modeling how these
choices interact with the flash cell and can be tweaked to improve speed, power
consumption and reliability. You can get a flavor of this in my article -
and DSP ECC IP.
In a conversation with
NVMdurance's CEO -
earlier this year (April 2016) almost the first thing I did was try to relate
and place the work they were doing within the simple frameworks I'd written
So I asked him how similar it was to something which I
wrote a long article about in
April 2012 - when
SMART announced a
range of SandForce
driven SSDs which had 5x higher endurance - while using exactly the
same industry controller - but using magic tuning numbers which they had
learned from analyzing the adaptive settings from their own adaptive controller
Pearse said - yes - he knew that work. And what NVMdurance was
doing was the same type of thing.
He said that some leading companies
which had the flash talent had done similar things in their proprietary SSDs
Pearse told me that as the complexity of flash increased -
with more layers and TLC - it was becoming harder for designers to manually
(or using human expertise) guarantee they were choosing the optimum magic
numbers - because there were now so many variables involved.
said that what was different about NVMdurance was that they were delivering the
magic numbers based on characterising a sample of typically 100 devices and then
performing machine based simulations to see which numbers would work best -
while also using a multi-stage life cycle model - which was designed to use
different tuning after a fractional amount of the expected endurance had been
As far as he knew from his conversations with memory companies -
no-one else had made the same kinds of investments in this machine intensive
modeling - and that was the key difference - because NVMdurance had a proven
process for delivering good tuning numbers over a variety of memory generations
I hoped at the time that someone would write a paper saying
more about it. Tom
Coughlin has done that.
learning enables longer life high capacity SSDs (pdf) - published this week
describes the background principles and operation of NVMdurance's pathfinder and
plotter software tools and shows you how NVMdurance have tackled this complex
tuning problem to deliver a software delivered IP which can give endurance
results which are similar to adaptive adaptive R/W controllers but which don't
need such expensive processors or such complex run-time firmware. ...read
the article (pdf)
Toshiba samples 64 layer 3D TLC
Editor:- July 27,
2016 - Toshiba
today said it is sampling 64 layer 3D TLC flash in a 32GB device and plans
production in the first half of 2017.
Editor's comments:- You can judge
the progress on this technology by the fact that in
March 2015 -
Toshiba was sampling 48 layer MLC.
Seagate unveils 2TB enterprise M.2
Editor:- July 26,
2016 - Seagate
a 2TB M.2 NVMe PCIe SSD
aimed at the enterprise market. The Nytro XM1440 M.2 - which will ship in
November 2016 - will offer 30K IOPS / Watt.
Memory1... if you want it - you can have it
war phase of DIMM wars has now ended
Editor:- July 20, 2016 - Diablo Technologies
the production release and volume availability of its Memory1 128GB DDR4
system memory module.
Editor's comments:- Diablo's Memory1 unveiling
in August 2015
was part of a wave of similar announcements last summer. I commented on its
significance and listed the other contenders which emerged in an article -
on SSD DIMM wars technology?
The phoney war phase of SCM DIMM wars
has now ended with volume shipments of Memory1 (flash as DRAM).
now time for competitors who aren't yet shipping products in volume to start
publishing full technical specifications for their own alternative memory
products (if they dare) and for investment oriented stock-price puffing
announcements about DIMM wars productsto cease and be replaced by papers,
ads aimed at engineers.
|What happened before?
|seeking the inner SSD
|9 years ago - in
Violin Memory launched an assault on the high end RAM SSD acceleration market
with general availability of the highest density rackmount memory system.
|I think it's not too strong
to say that the enterprise PCIe SSD market (as we once knew it) has exploded and
fragmented into many different directions.|
|what's changed in enterprise