AllFlash

All Flash Debate
The future of the datacenter is all-flash. Or is it? Join our expert panel and share your opinions.
   5 years ago
#allflashAll Flash Impact Analyst and crowd perspective on All Flash impact to cost and performance
Dave Vellante
Q1. As the world moves toward a digital economy is storage infrastructure still relevant to the question of data value?
Eric Stouffer
data has to be stored somewhere, so absolutely yes
Randy Arseneau
Absolutely! The value of data as an enterprise asset has never been higher, and it's growing geometrically to boot.
Storage Alchemist
at the end of the day for the consumer, maybe not - they may not pay attention to it but for the people hosting the data - YES
Marc Farley
data is really the only asset worth worrying about - or should be
Randy Arseneau
@esstouffer Yeah - gotta live somewhere.
Peter Burris
Data is the new asset -- the main asset -- for the "digital economy."
John Furrier
latency will be the big driver that answers this question; so yes it will always be a storage issue
Kris Tuttle
Seems like consensus here is "yes"
Peter Burris
It ain't digital if it ain't data.
John Furrier
The questions is what is storage?
Drew Schlussel
not all data is valuable; flash is overkill unless you're working in (near) realtime
I am John White
I am not sold that we need to keep everything. I think some storage is used for junk data making it not so valuable
Storage Alchemist
its also about the competitive business process - its the data, and how it is analyzed etc...
Dave Vellante
Drew - we'll be discussing that in some depth today
Yaron Haviv
yes, but storage need to evolve to allow extracting more value from the data
David Floyer
Yes - how quickly data can be accessed and stored is a significant factor in data value - in general the fresher the data the greater the value
Peter Burris
The relationship between data value and storage is pretty simple: Does the storage make it harder or easier to exploit the data.
Marc Farley
there is no need to keep everything, but if we are looking for needles in haystacks: ala analytics, when do we know data is obsolete?
Storage Alchemist
@johna_white How do we know though - "One mans trash..."
Randy Arseneau
Data = currency of differentiation IMO. Those who exploit if fully and efficiently win.
David R. Klauser
beyond "still" relevant, having focused on converged solutions past few years, it is and must become increasingly relevant and worthy technically and financially of a stack that commands it's own decisions
Drew Schlussel
@skenniston If you don't know, why are you collecting the data in the first place?
Storage Godfather (HPEStorageGuy)
It's always been about the data - storage is just a place for it. I'm kind of partial to IKEA Storage myself.
Jorge Hernandez
The challenge will be to be able understand data on time, and not when is too late and meaningless.
John Chen
The analysis of data is key to monetization, so the easy of access and scaling at lower costs is key.
Mike Jochimsen
The problem is the value of data is sometimes not known in real time.
Peter Burris
For example, as David Floyer says, being able to copy data and make the same base available to many applications at the same time dramatically increases the value of any data.
Eric Stouffer
maybe a more provocative Q is what will that infrastructure look like
Big J Maestre
the idea that storage isn't relevant is ridiculous. IDC shows that more TB's & more units are being shipped than ever before. There are more storage vendors than ever as well. Do NOT confuse shrinking $/GB w/shrinking value. Block Storage is still key.
Peter Burris
That makes flash a better "data value option" than most other types of storage, other things being equal.
Craig Nunes
Good point on latency--that all depends where you measure it...at the workload or across the network on an array.
Dmitry Golubev
efficient storage is the key to extract value from data, eventually it comes to not erasing data at all when PB/$ is low.
Steve Hunsaker
CPU & RAM are solving latency issues...almost making the type of disk performance irrelevant
Marc Farley
latency is not always key. cold data doesn't need low latency
Steve Hunsaker
@gofarley Very true. Auto-tiering then comes into play.
Peter Burris
Data value is driven by context, and context is time dependent. Data that is not valuable now, may be enormously valuable at some other time.
Dave Vellante
Q6. Some say that all flash puts faster media in the same old architecture - is this fair? Why is this relevant?
Brian Biles
If it's still on a SAN, it's still swimming upstream against ever cheaper, more abundant host memory types
Matt McDonough
yes in most cases the I/O stack is the same, you still have controllers in line of the datapath. You need to remove that stuff to take full advantage of the flash you bought.
Randy Arseneau
My '72 VW Beetle will go faster with a Cosworth engine. Great until I have to turn...
I am John White
Very fair and very relevant. Ask a storage company for more bandwidth capacity and they tell you to throw 10Gb ports at it. Doesn't last forever
Storage Alchemist
in a lot of cases it was time to market and putting the Porsche engine in a SmartCar - architecture does matter http://www.infinidat...
Randy Arseneau
We started with the architecture, then solved for media.
David Floyer
Sure it was fair for the early all flash arrays. They have matured a lot since. Big impact will be Server SAN. Best SPEC1 of 5 million achieved by Datacore with 25 microsecond access time and 1/4 of the cost of traditional arrays.
Drew Schlussel
Yes, because nobody wants to reframe the problem. The architecture was built to accommodate the fastest component (CPU/on-board RAM), not the slowest (disk). Now that changes with #allflash and so does the architecture...
Dave Vellante
So what happens to storage architectures? Do they change dramatically? Do apps have to be re-written?
David R. Klauser
putting all flash into traditional architecture will prove limiting at scale, the vendors who optimize "old hardware" w/ optimized algorithms will optimize the potential of flash
Randy Arseneau
Yes. And yes. I/O path has to be simplified and optimized.
Rodrigo Gazzaneo
All Flash on SAN is about accelerating the existing stack. A valid use case.
Storage Alchemist
new technologies and newer strategies have evolved as we have gotten smarter in the storage business - times have changed - need a better architecture
Randy Arseneau
@vGazza It is if you want to go broke... :)
Matt McDonough
in order to get the very lowest latency, apps need to be re-written, but that is the top 5%. Look at what DSSD is doing. Lots can be done even with standard block access.
Rodrigo Gazzaneo
@johna_white thus the emerging interest on #NVMe fabrics
I am John White
With the apps already being written in greater horizontal distribution then before it starts to solve the problem naturally if you scale the underlying architecture the same way. Hence, HCI makes a lot of sense.
Eric Stouffer
i think it is mostly true, so "fair" is subjective and probably irrelevant; until something comes along to change the physics of moving data to/from storage
Randy Arseneau
Containerization will eventually compress the stack dramatically.
Satinder Sharma
Fair as long as old architecture can guarantee low latency all the times at a low $/IOPs and a competitive $/GB
Yaron Haviv
NVMe over fabric, adding steroids on an old (SAN) architecture, did anyone see a cloud provider using SAN ? they all use DAS with scale-out at the data management layer
Dave Vellante
@vGazza Can you elaborate on NVME?
Storage Godfather (HPEStorageGuy)
Fair to a large degree - when will the compute model change to make use of new technologies like photonics and memory fabrics? SSD is a bandaid for the next compute model.
Aaron Skogsberg
>> flash in trad architectures "will work" however there are tradeoffs in terms of incorporating efficiency techniques that customers appreciate
Kris Tuttle
@CalvinZito Interesting comment "SSD is a band-aid for the next compute model."
Marc Farley
all-flash is the precursor to what will happen when NVMe is utilized. We have never seen anything like this in our industry before.
Randy Arseneau
@CalvinZito Good point. Although again, industry will be fab capacity constrained for some years to come...
Rodrigo Gazzaneo
@dorkninja it is a use case story today. But flash is evolving fast.
Kris Tuttle
Why is NVME such a big deal? Just faster?
Storage Godfather (HPEStorageGuy)
@Heuristocrat Because SCSI wasn't designed for low latency.
Drew Schlussel
@dorkninja That's by design. Maintains the profit margin to support additional capital investment later...
Rodrigo Gazzaneo
NVMe redesigns the stack from legacy serial disk access protocols (SCSI)
Keith Townsend - Light will overcome darkness
From EMC with DSSD5 to the lower end of the market with ScaleComputing, vendors are addressing the cruft associated with layering legacy on top of FLASH. But fair question.
Big J Maestre
The problem is w/your question. Storage architectures just work. Scale is managed better, performance is better, bandwidth is better, latency is lower, etc. Replacing Disk with SSD just makes all those points better. It's an inarguable fact.
Rob Peglar
@Heuristocrat NVMe is important - a much lighter-weight, CPU-and-memory protocol, replacing SCSI in which the target is in control of the operations. NVMe is much more balanced and capable.
Marc Farley
its memory to memory transfer, that can use SSDs as memory, cache or disk. It runs over any physical network (remote) finally it enables up-stack access to these memories.
Yaron Haviv
wrote a post on over emphasis of low-level storage (& NVMe) and ignoring the new app models & arch https://www.linkedin...
David R. Klauser
it can optimize SSD w/ PCIe
Rodrigo Gazzaneo
@Heuristocrat NVMe allows for other application models that can use flash as memory and not emulated disk
Aaron Skogsberg
@vGazza >> the question becomes with this redesign how are vendors helping this transition? Should not require a forklift process which most vendors will require.
Randy Arseneau
Yes, but capital investment in fab is higher leverage, higher risk. Architecture bets are high stakes.
Mike Riley
Apps are being rewritten - the rise of 3rd platform is far more disruptive than flash. Flash didn't change biz models. AWS did.
Drew Schlussel
YES! YES! YES! The revolution really begins when the "classic", core apps are rewritten for the new frameworks AND new hardware architectures.
Rodrigo Gazzaneo
@DweebiaK the forklift is happening at appl level. Modern apps, modern stack. Legacy apps, legacy stacks.
Rich Pappas
@MSR11 Great point, the underlying infrastructure has to comprehend that business apps are changing faster than ever
Dave Vellante
@MSR11 I like that angle mike - agree
Aaron Skogsberg
@vGazza >> it actually doesn't have to. Modernization at the foundation enables transformation higher in the stack. No need to migrate the data. The data is the array not the physical entity.
Rich Pappas
@skenniston I think the architecture on prem will need to be more versatile, everything else goes to the cloud
Aaron Skogsberg
@vGazza >> the key here is #NoForklift. Get rid of the legacy process & procedures and help organizations embrace technology that aligns to the changes in their business even through generations. #Simple #Invisible #Automated
Rodrigo Gazzaneo
@DweebiaK it is the app that accesses the data that cares about the stack
Aaron Skogsberg
@rpappas >> Concur --> Evergreen!
Rodrigo Gazzaneo
@DweebiaK that works if you are providing faster storage for your legacy app. Developers want more.
Aaron Skogsberg
@vGazza >> The foundation of the stack should be #invisible. Organizations want agility not the burden that comes from traditional inf stacks w/ legacy processes, procedures, etc
Rodrigo Gazzaneo
@DweebiaK invisible is for legacy. Apps that drive IO. Modern apps will drive persistence via API.