"It's so Fast"...Flash Storage 101

Warning! This is a quasi tutorial blog post (or introductory 101 primer) for those wanting to know a little bit more about "Flash Storage" and why its popping up everywhere these days in Enterprise IT circles and general purpose computing. It gets a bit involved, detailed and tedious -- but I tried to keep it high level and informative just the same. If you wade through this to the end, you'll hopefully get a clearer understanding of what Flash storage is; how it's typically 'packaged' or used and -- most importantly -- how to get the most from it. At the very least you'll get an idea of what all the buzz is about...and how Flash Storage is transforming datacenters and leading today's "Digital Transformation" technology curve...

?Understanding Flash Storage and Getting the Most Out of It

Nowadays, storage specialists and IT generalists alike are well familiar with “flash” storage and the advantages a flash storage device has over “traditional spinning magnetic media” – i.e., those tried and proven hard disk drives (HDDs) we all know and rely on (and sometimes loathe...when we get a hard disk/head crash). 

Let's face it. Hard disk drives with magnetic spinning media have been around for a long time. All HDDs today can trace their roots (and technology) to this early 'Winchester drive' that IBM debuted decades ago.

Above is an early 4-platter Winchester Hard Disk Drive (looks like a 5 incher) circa 1970s...tried and true spinning disk storage media technology (albeit, dated and probably with only 5MB raw capacity)

You can think of HDDs as being the audio equivalent of vinyl records or cassette tape -- albeit, of the 'writable' digital magnetic variety (instead of "analog"). Flash storage on the other had, would be like those groundbreaking solid state iPods from Apple...or simply your smart phone. Of course, the more 'memory' (or Flash storage) the quicker you'll end up filling it... :-)

Where as HDDs store binary bits (that form bytes...and "data") on their recording media magnetically, Flash storage (as with RAM) stores data (i.e., "Write I/Os") by setting '0' or '1' binary bit data using very small voltages...whose electrical bit values persist even when powered down.

Moreover, Flash storage nowadays is not like the older "write once, read many" (i.e, WORM drives) CDROM disks, optical or early Flash devices you might have heard about. Contemporary Flash storage devices handle many/many Write and Read operations equally well; with reliability and mean time before failures (MTBFs) up there with HDDs...or maybe even surpassing them. Granted, Write operations do take a bit longer (latency wise) than Read I/Os and everytime voltage is applied to Flash Cells for Writes there is a bit of 'degration' with that cell's insulator material. But nowadays any "Write penalties" with Flash are fairly negligible...and getting better all the time.

Since roughly the 2010 time-frame, Flash storage adoption rates have steadily increased as $$/GB costs continually decreased. And since 2012, this adoption rate across the IT universe, enterprise data centers and even mobile users has been skyrocketing -- exceeding most predictions made only a couple of years ago.  If the trend continues, Gartner estimates that by 2018 Flash $$/GB costs will be on par with standard 7200 rpm SATA HDD $$/GB costs. Turns out Flash $$/GB high raw capacities will have already achieved cost parity with 15K rpm highest performing HDDs by the end of 2016 – or so some notable analysts predict. 

The Flash Factor

If you've been even modestly keeping up with various IT industry readings, I'm assuming you’ve come across the terms “Flash storage”, “SSD” (Semiconductor Storage Device), “Flash Cards”. Or maybe even “Flash Caching”, “Storage Pooling” or “Storage Tiering”.  These terms (and use cases) frequently come up when today’s “Digital Transformation”; commodity based “Software Defined Storage” or “Hyper converged Storage Systems” contemporary IT terms come up. [Feel free to search up these terms "on our omniscient and trustworthy internet".]

Turns out, Flash has emerged from the bleeding edge to a well-established leading edge key component for these new, emerging technologies and computing use cases. And Flash storage devices will continue to be is a significant change agent as a key enabler in transforming the IT datacenter, Cloud and computing in general for optimized handling of modern applications, new storage deployment trends and demanding high performance workflows.  

What’s behind the widespread usage of Flash storage in Datacenters? Quite simply, high performance, affordability, reliability and robust data services (or SW) that leverage Flash storage advantages and value proposition.

Already the most tangible positive impact of Flash storage deployment has been its role as a major catalyst in the emergence and adoption of the aforementioned industry trend towards adoption of industry standard hardware (i.e., “commodity”) based software defined storage (SDS) ‘server SAN’s and hyper converged infrastructure (HCI) storage platforms that eliminate the need for specialized, vendor specific proprietary traditional storage hardware and software systems. These ‘disruptive’ – but complementary – storage technologies are changing the Enterprise IT landscape, along with “The Cloud”, mobile computing and emerging “Internet of Everything”.

“Moving at Tron Speed…or Close to It”

Flash storage technology and Flash chip semiconductor packages cast with silicon wafers have been around for a long time. Flash storage is a very close relative to higher performing (and more expensive) solid state (i.e., silicon) memory chips. Nonvolatile memory NAND based silicon chips; whereby the actual data trons and bits written to it are retained even after shutdown or power disruption. But over the past ten years, Flash storage – previously reserved for ‘special’ high performance I/O use cases on pricy high end servers (i.e., boot drives or “extremely hot” data caching) – has become more economically viable and attractive for a much wider variety of storage use cases.  So much so, 2.5-inch and 3.5-inch Flash drive devices (packaged like regular HDDs) are not uncommon in commercial servers, desktop PCs and laptops.

In parallel with falling costs and end user pricing, overall maximum raw storage capacities of various Flash storage products have also increased dramatically. 16, 32 and – now 64TB raw high-end capacities modules or "blades" are not uncommon. So this favorable price reduction in tandem with capacity increases (and better reliability, longer lifecycles and faster Write performance ) is driving higher penetration rates of Flash chip based SSDs that will soon rival (or likely surpass) HDD market/deployment share numbers. In addition, the aforementioned improved data reliability and longer lifecycle spans due to over-provisioning of Flash cells, load balancing internal firmware/software algorithms and longer lasting multi-layer cell technology (MLCs) have also improved to the point where MTBFs and lifecycles are as good – if not better – than standard spinning media.  

But it’s not just about falling prices, improved performance (especially for Writes), higher reliability and/or longer lifecycles. Flash storage is…well….FAST! With no moving electro mechanical parts…

Flash Technology?

Flash semiconductor chips (typically a BGA package) are the key ingredient/component behind all SSDs, PCIe Flash Cards, Flash Blades, Modules and All Flash Arrays (AFAs) Flash storage packaged devices. Individual Flash silicon chip die (or wafers) make up a discrete/individual Flash chip package comprised of cells set by voltage (binary 0s or 1s) able to store a certain amount of raw data in either single layer cell (SLC) or multi-layer cell (MLC) Flash architectures.

Printed Circuit Board (PCB) Mounted with NAND Flash Die (Chip Packages)

Simply stated, Flash is "slower" (and cheaper) non-volatile memory (NVM) based with those electrically charged memory cells (single layer SLCs or multi-layer MLCs) cast in silicon and NAND circuitry. Think of it as a cousin (albeit, slower and cheaper) to your smart device, PC or server CPU’s companion RAM (random access memory). SLC based Flash storage typically are even faster (and pricier) than MLC based Flash cells. The MLC advantage? Higher raw storage capacities with lower pricing -- but MLCs are still lightening fast compared to magnetic spinning media (i.e., disk platter) based storage devices. Plus, MLCs are available for either consumer or commercial grade quality (i.e., eMLCs) applications, and packaged and priced accordingly.

What’s the first thing most of us look at in order to jazz up our laptop’s CPU throughput performance? More memory. Larger memory capacities and faster clock speeds increase I/O throughput and reduce I/O latency – and keep I/O request traffic/responses closest to the host CPU. Similarly, depending on how it’s packaged and where it’s installed, Flash storage is second only to DRAM performance and their low I/O latency.

Those of you who have laptops with an SSD flash drive(s) already know how much faster they are over standard drives.  Much improved, higher performance…in terms of I/O operations per second (IOPS), higher bandwidth (MB/sec) and much lower response times (Latency). 

Below, a contemporary electro-mechanical Magnetic "Spinning Media" Hard Disk Drive with movable Write/Read heads and multiple disk media 'platters' (HDD):    

 

Compared to a no-moving-parts Semiconductor "Flash" Solid State Drive (2.5- or 3.5-inch) comprised of multiple Flash chip packages (SSD) 

Which do you think is faster?

“Flash is Fast….Flash is Cool” 

Per above, Flash is truly and rapidly changing the storage landscape. In fact, IDC has proclaimed that 2016 is “The year of Flash”. The graph below from Wikibon illustrates the steadily decreasing cost of SSDs (and by implication “Flash” storage chips) and their projections beyond 2016.

Decreasing SSD Costs -- Chasing Capacity HDD Costs/TB (Wikibon)

Once quite expensive, higher performing Flash storage continues to enjoy decreasing prices...and that trend is expected to continue. This trend -- plus increasing capacities and longer 'life'/durability of Flash cells -- is driving up market adoption across the IT storage spectrum.

Flash Storage Key Value Props

Net/net? Flash storage in general and SSDs in particular are enjoying an accelerated market adoption rate as they keep approaching HDDs $$/GB and capacity numbers. But in addition to pricing, Flash storage attractiveness and market acceptance is also driven by the following:

1.     Flash is ‘just silicon and trons’ with discrete address locations (or cells) for bits & bytes of data – with no moving parts – so it’s optimized for random Read I/Os

2.     Flash is fast – much much faster than any/all HDDs in terms of IOPS, throughput and reduced latency of I/O traffic requests

3.     Flash provides higher storage form factor densities and superior packaging advantages

(ex., less cooling, electricity/power and rack space overhead compared to spinning disks)

4.     Flash lifecycles and reliability (99.9999% for Dell EMC’s XtremIO AFA, for example) have improved dramatically, along with ever increasing raw storage capacities (i.e., 32TB 2.5-inch SSDs are the latest/greatest versions)

 

Rating Storage Device (and Flash) I/O Performance

So just how is storage I/O “performance” measured and compared?  By these three key, well accepted storage industry metrics (and jargon):

• IOPS  -- number of Read or Write operations each storage device or overall host system can process in/out  (i.e., I/Os per second). You can think of this sort like being a CPU's clock frequency or instructions per second 'MPH. speed.
• Bandwidth or Throughput – overall maximum data volume or rate which bits or bytes can flow to/from the host CPUs for application I/O requests (i.e., MB/sec or GB/sec). Since this pertains to a flow rate, you can think of this as roughly being analogous to a water pipe's diameter...the wider the diameter, the quicker your downloads.
• Latency Response – actual time it takes (usually in mili or micro seconds) to complete a discrete I/O request loop (i.e., millisecond or micro seconds, usually averaged). Some people think this is overall 'user response'...(as is time you see results show up on your screen. But be careful! I/O latency and device seek, transfer and response times can be different than your overall PC (system) latency or even your UI screen display results.

While a typical HDD might deliver @ 160 – 180 IOPS at the individual drive level, a typical enterprise grade SSD can produce around 9,500 IOPS per device. PCIe I/O Flash cards can exceed 150,000 IOPS at the card level -- orders of magnitude faster than the faster 15K RPM high performance HDDs currently available. 

Of course, these rough comparison numbers are at the device level and, admittedly, I've over simplified things here. Actually, the above performance measures (for any storage device) are determined by the actual data workloads and "I/O characteristics". Per my opening comments above, generally speaking individual data Write operations take longer than Read operations with any Flash storage device/type. So usually any of the above metric test results are expressed in workload percentage/ratios of "Reads vs Writes" (i.e., 0/100, 50/50 or 100/0). So read that 'fine print' if you or your staff are reviewing or evaluating Flash storage products -- regardless of how it's packaged and deployed. [Some vendors are more 'optimistic' than others in presenting various specs]

Bear in mind, too, that "your mileage will vary” according to your actual host system running the show and where/how your Flash devices are attached to. But regardless of how it’s packaged and where it’s installed on a host CPU/server or PC or shared storage cluster, Flash is the undisputed king of storage performance by any/all of the above performance measures. It will always be faster than HDDs, even when said Flash is not optimally installed or controlled.

Why is this noteworthy and important to Enterprise IT datacenter administrators?Deployment of Flash storage has the advantageous ability to do move data in/out of storage and back to the host CPUs (or VMs and apps) much faster…with less “spindles” (i.e., "drives") or the need to performance tweak them (i.e., firmware upgrades or 'short-stroking' of said drives with optimized data record placement on the platter with fastest head seek/transfer rates). 

These favorable Flash characteristics translate (typically) into fewer storage devices; the ability to maximize host CPU(s) resources (less time spent waiting for a completed I/O and/or more efficient use of CPU cores); better handling of non-sequential, randomized data (in step with today’s modern applications and computing platforms); higher storage densities and less rack space/datacenter overhead. All this which in turn translates into improved operational dynamic flexibility, data storage efficiency and lower TCO (total cost of ownership for the whole solution; i.e, capital expenditure and operating expenses, etc).  Flash also key for enabling new software defined storage platforms (SDS), hyper converged (HCI) platforms and faster handling of unstructured file, object and/or containerized data and that much touted “Cloud”.

Turns out not only “Flash is fast and cool”….”Flash is cash” too. This is fortuitous since Enterprise IT datacenter administrators these days want to (or must) reduce TCO through reduced capital expenditures (CAPEX) and operational expenditures (OPEX). 

A high priority (and means to lower TCO) is to reduce platform and vendor lock-in by leveraging as much as possible industry standard hardware and software defined ‘everything’ solution and/or virtualization architectures. Additionally, since Flash storage is generally 'faster, more powerful and flexible' than traditional HDDs, IT administrators can therefore consolidate application/user workflows from many, disparate storage systems (aka, 'storage silos') to fewer -- or even a single -- Flash based storage platforms. Such workflow consolidation is a whole separate subject and movement going on today in the storage industry...i.e., doing more with less (resources).

Flash Packaging and Deployment Models

Generally, the main types of Flash storage are packaged in SSD (semiconductor storage device) industry standard form factor drives or PCIe I/O accelerator “Flash cards” that mount in standard host PCIe (i.e., PCI Express high speed host serial interconnect) card adapter slots. Flash storage is also available in custom ‘blade’ configurations (usually custom) as well as for drop in replacements (of HDDs) for All Flash Array (AFA) external storage array boxes. 

How best to actually use or deploy Flash Storage? It depends, of course, on what the system (and client application) objectives are. Basically, there are four different ‘models’ or approaches:

1.     As a direct attach (DAS) “I/O accelerator card” or local storage drive for host CPUs – i.e, PCIe I/O Flash Cards and/or SSDs as a ‘local storage device’, respectively.

A PCIe I/O ‘slot’ Flash Card --         

• 'Flash cards' attached directly to host servers or PCs (custom or industry proprietary product)
• Inserted closest to the host CPU (after RAM) in a high-speed PCIe serial interconnect slot; can be used in conjunction with standard SSDs and HDDs 'downstream' on the host computer
• Used/dedicated to host CPU (i.e., not ‘shared’ on a network), same as 'local' SSDs and/or HHDs

Data is stored locally, data protection on a cluster typically requires software installation on participating hosts to copy and share data (i.e., a SDS or high availability (HA) dual controller storage array architecture)

2.     As a “pure” Flash storage box – i.e, external (or direct attached) “All Flash Arrays” (AFAs) for SAN or NAS data comprised (usually) of SSDs or (increasingly) proprietary/custom Flash modules or “blades”.

                                         

                        

  

Above: An external AFA 'brick' from Dell EMC (XtremIO) comprised of all SSDs.

Left: A Flash Module....or "blade"

·     

•  Contains all SSDs (pure) or flash modules/’blades’; shared cluster storage array ( new ones like above Dell EMC’s XtremIO are purpose built for Flash)
• Consistent and extremely high IOPS ‘appliance’ with low latency (i.e., 1ms)

                                         

3.     As part of a host or ‘hybrid storage array’ Flash data cache – i.e., Read, Write Through or Write Around data cache in front (upstream) of ‘traditional spinning media’ storage drives for faster access of frequency accessed or most recently written/read ‘hot data’ typically shared on cluster via dedicated Fibre Channel or Ethernet storage network.

A Dell EMC Hybrid Storage Area Network External Array (VNX)

• Consists of both SSDs and HDDs in same chassis/installation
• @ 2 – 5% SSDS vs total HDD raw capacity for hot data caching and/or creation of high performance storage tier (pool)
• Hybrid arrays can be block, file or object data based (or combination)…shared storage accessed by dedicated storage network (i.e., ‘fabric’) for clusters

4.     As a separate storage tier (or pool) or as part of a mixture of Flash/SSDs and ‘traditional’ capacity HDDs ‘hybrid pool’ formed DAS storage devices (on host servers) or from above shared/external hybrid storage arrays.  

Rather than relying exclusively on a data cache on the host client/compute server (i.e., node) or dedicated storage only node (or array); users can actually create and designate separate storage tiers as they like to handle particular application, use case or workload I/O performance needs. Flexibility…with high performance. 

• For SDS and/or hyper converged virtualized storage, tiering can be used for industry standard server appliances
• Flash can be deployed as all flash tier/layer or for multiple tiers consisting of a local RAM, PCIe I/O flash cards; SSDs and HDDs (mixed or separate)
• Per above #3 hybrid ‘traditional storage’ array model, mixed storage tiers (or pools) can also be comprised of Flash only, or can also be a blend of high performance SSDs and lower cost/higher capacity HDDs

Storage tiering (or Flash tiering) is really great for applying storage performance or capacity (or both) when and where you need to maximum application and system I/O performance characteristics – tuned for hot, warm, cold, frozen or petrified data storage as required.

Note: Caching is not mandatory or necessary for 'traditional' hybrid storage arrays or the newer software defined storage (SDS) virtualized platforms installed on industry standard 'commodity' server hardware (ex., a Dell PowerEdge server) or pre-packaged hyper converged 'server SAN' vendor appliances, but is not excluded either. 

Tiering?

Like caching, the concept of storage pooling or tiering is also not a new one or exclusive to Flash storage. Tiering has been around a long time. It so happens, however, that Storage tiering -- with Flash as the highest performance storage tier for hot randomized or sequential data -- is a fundamental concept and deployment model for high performing, efficient virtualized software defined storage system (SDS) bundled hardware storage server appliances or ‘pure software’ DiY deployments. 

The diagram below courtesy of SanDisk (a large manufacturer of Flash based storage devices) depicts multiple storage tiers, including DRAM and Flash devices. It also illustrates the evolution from legacy storage devices to the emerging/newer flash centric storage deployment models; and how they are typically applied. In the illustration, ‘Tier 0’ represents that highest/fastest performing storage tier while ‘Tier 3’ (or below) typically is comprised of lower cost, higher capacity storage for near line ‘warm’ or archived ‘cold’ data use case scenarios. 

What's the tiering advantage? Greater dynamic flexibility, capacity, cost control and ‘IOPS, throughput and low latency’ -- when and where you need it. With separate storage tiers selected for high performance, high capacity or lower $$/MB cost usable storage, users can shape their respective storage tiers (and data copy/protection policies) using automated management tools available with the SDS cluster or 'traditional storage' array platform.

Really good, flexible systems -- such as Dell EMC's ScaleIO SDS architecture based storage solutions -- allow users to manually create, modify and/or tweak different storage tiers easily at will, and according to application, data 'temperature'/time value, client multi-tenancy (especially for private or public clouds) or general user SLA (service level agreement) or overall data i/O accessibility needs.

In closing, Flash storage devices essentially can be used for high speed "local" client or data caching. They also add yet another high level performance storage tier to the IT administrators tool box and wheel house. Caching is a cool tool too, and will always be around. But think of caching as a subset of sorts to the overall Flash storage value add story...as well as being somewhat of a complementary mechanism (or subset) to storage tiering practices in general. Oh...and did I mention really great hyper converged and/or software defined storage solutions (again, a plug for Dell EMC's ScaleIO) can seamlessly use "The Cloud" (private, hybrid or public) as yet another storage tier for really cold, low-cost data storage? A lot of companies and data centers these days (big or small) are using Cloud storage in this manner to archive and back-up their data "off premises".

For additional tutorial blog info on "Caching versus Tiering", I've posted another blog here for your reading 'pleasure' and information. Hopefully, this blog now has properly addressed the importance of Flash storage; how it's transforming the modern datacenter and how its packaged, deployed and used -- in addition to familiarizing you with all the tech jargon associated with these new emerging IT storage technologies!

For more info on Dell EMC storage product info, please follow @DellEMCstorage; @DellEMCScaleIO or @DellEMC_CI (converged) on twitter. Or open the link below: