Can memory do more? - article on StorageSearch.com


leading the way to the new storage frontier	.....


the SSD design heresies	..


nvm	....


DIMM wars	....


top SSD companies	..

where are we heading with memory intensive systems and software?

there's more to upcoming change in SSD than DIMM wars

by Zsolt Kerekes, editor - StorageSearch.com - June 17, 2016

The SSD market is moving into a new phase at the core of which is changed thinking about the role of memory and storage and software.

In past such transitions (for example the RAM SSD to flash SSD transition in enterprise SSDs followed by years of successively different types of flash) it was much easier to predict what problems needed to be solved to enable such memory adoptions in different markets - because they were all about how do you make reliable enough systems out of new, higher density memories which have intrinsically poorer data reliability than the devices which they will replace?

And once SSD architects understood flash well enough - the direction of successor challenges was predictable...

Another denser type of flash.

These presented detailed problems for controller designers and memory makers - but didn't change the inevitable laws of memory succession.

The background business principles for adopting such products were also clear and had remained the same for 10 years.

what's different now?

What we've got now is a new SSD market alchemy melting pot - which I alluded to in last year's article - the SSD Bookmarks - why we need to begin thinking again.

At its simplest - the factors now at play are these:-

When all storage is made from memories the dividing line between storage and memory is much more fluid than it has been before.

If you have enough flash memory chips you can build an SSD.

And if you start with that very same flash SSD - you can make it behave like core memory - DRAM - which is an entirely different type of memory than that used inside the SSD.

Interesting questions are:- How much memory can we replace with such an SSD? Where does it work best? What are the limits?

You could say - using an analogy from thermodynamics - that storage has lower data entropy than memory - because you can do more with raw memory than with raw storage. The heat pumps which change the roles of one memory package to another are coming from the software market.

And just as flash SSDs needed characteristics like endurance and DWPD to help us understand their limitations perhaps virtual RAM needs entirely new latency spectrum based specifications too...

There are more type of memory entering the storage / memory mix than ever before.

It's not just flash (which itself has got more dimensions of choice than before) but alternative nvms too.

What do want our memory systems to do for us? And how do we want them to behave?

In the past all such efforts focused on making memory systems more reliable and compatible with past software.

An important emerging trend has been to demand much more from memory systems, and to offload data compute and other functions where ever it makes more sense from an energy and performance point of view to perform such functions inside the memory array.

Who are the companies leading these changes?

Which are the products and technologies which can be regarded as iconic in their classes?

From the marketing point of view one of the loudest technology categories which falls within the above market changes has been the SSD SCM DIMM wars market. If you click on that link you'll see a long list of companies which have made promises in that direction (Micron, Intel, Diablo and many more).

But that's only part of the story of the rewriting of the memory plot.

In the past 3 years there have also been growing developments in the area of making memory systems more capable, more manageable by software and more flexible.

Here are some of the companies and strands of thinking involved.

in-situ processing

NxGn - in-situ processing in flash SSDs.

Altera - DRAM coresident FPGA architecture to enable CPU offload of IO and compute.

Radian Memory - providing offload functions in flash SSDs

software managed memory

Plexistor - manages the convergence of memory and storage in big systems from a software approach it calls Software Defined Memory.

Marvell - has a firmware and controller associated architecture called Final-Level Cache - which provides flash as DRAM functionality - which can deliver higher performance and lower electrical power.

new things in DRAM

Symbolic IO - has used data coding techniques embedded at CPU level to dynamically extract more usable capacity and performance from in all the memory types within its enterprise systems from DRAM upwards.

Encrip - has demonstrated a chip level IP which can be retrofitted to most conventional DRAM chip designs - to provide 3 levels of internal coding instead of 2 (binary). The IP can deliver more usable capacity and higher performance - or it can be used to save power by reducing the amount of raw DRAM needed in systems.

DRAM fabrics

No matter how much memory (of whatever type) you can place in a rack it's never enough for everyone. A3CUBE has been at the forefront of companies pioneering low latency DRAM fabrics based on PCIe connected servers.

related articles

DRAM's indeterminate latencies

what's RAM really? - RAM in an SSD context

DIMM wars in SSD servers - the Memory1 salvo

are you ready to rethink enterprise DRAM architecture?

a beginner's guide to semiconductor memory boom bust cycles

SSD ideas which changed in 2014 - first appearance of in-situ SSD processing

SSD ideas which changed in 2015 - retiring and retiering enterprise DRAM

PS - One of the things which makes me laugh when I see big systems companies pouring money into born again "flash" marketing messages is that the enterprise SSD centric market existed before most of them had any SSD product lines. And solid state storage and big memory thinking - "SSD" for short - isn't tied to any specific type of memory.

It's the interaction of memory, data architecture, processing and software and their changing economic costs and application opportunities which are at the heart of the market.

Flash has played a big part in making SSD architectures affordable to be sure. But there's more to SSD thinking than flash.

....

....

Setting the scene for DIMM wars

The DRAM market's new clothes had long been invisible.

But the SSD market was too preoccupied with lower hanging storage fruit.

Now the secret is out and the effect of DIMM wars will brutal and swift and erode decades of collective wisdom about the shape of next generation memory.

latency loving reasons for fading out DRAM

....

SSD news
the SSD bookmarks
SSD controller news
DRAM news in SSD context
flash and nvm news in SSD context
after AFAs - whats the new flash box?
introducing Memory Defined Software
are we ready for infinitely faster RAM?

....

SSD market history

2003 - terabyte SSDs become commercially available

2007 - Fusion-io launched the ioDrive - a PCIe SSD

2008 - over 100 companies design SSDs

2010 - SSD market revenue exceeds $1 billion

2011 - year of Fusion-io's IPO

2012 - adaptive DSP ECC flash offered by over 10 companies

2013 - memory channel SSDs pose competitive challenge to PCIe

2014 - in-situ SSD processing promises new roles for SSDs

industry momentum grows for rethinking DRAM architecture

2015 - SCM SSD DIMM wars market exceeeds 10 companies

2016 - memory efficiency and speed redefined by coding from Symbolic IO

2017 - DRAM and flash shortages make alternative nvms look better

2018 - memory accelerators don't have to look like memories

There's a genuine characterization problem for the SCM industry which is:- what are the most useful metrics to judge tiered memory systems by?

is it realistic to talk about memory IOPS?

DRAM is neither as fast nor deterministic as you think

latency reasons for using less DRAM

how much flash is needed to replace all enterprise HDDs?

meet Ken and the enterprise SSD software event horizon

big idea #3

retiring and retiering enterprise DRAM

which includes a new value proposition for enterprise flash SSDs (flash as RAM) and presages a rebalancing of server memories - DRAM will shrink as a percentage of the physical RAM - which will also make it easier for emerging alternative memory types to be adopted by hardware architects and by systems software too.

What were the big SSD ideas of 2015?

...

Trying to understand the SSD market (based on what happened in the past in memory) will give you the wrong clues to business trends.

an SSD view of semiconductor memory boom-bust

...

Who's got all the answers to help understand how all the changes in the SSD market are coming together?

The answer is - no one and everyone and you too.

the SSD Bookmarks

With the ratification of the NVMf specification in June 2016, flash devices such as SSDs and storage arrays can now communicate over RDMA networks (such as RoCE or InfiniBand), delivering the same high performance, low latency benefits as local attached NVMe.

what were the big SSD ideas which emerged in 2016?

Would users buy an SSD just because it has an animal in the logo or label? If so - what animal should it be?

animal brands in the SSD market

Why can't SSD's true believers agree on a single shared vision for the future of solid state storage?

the SSD Heresies

Companies like Xitore and Netlist have been saying they want to get into the "flash as RAM in the DIMM form factor" market for a while now.

I haven't seen details from these expected competitors but my guess is that - unlike Diablo's product - which leverages the DRAM which is already in other DIMM sockets in the same motherboard - that some of the later contestants in this market will take the approach of placing everything needed to provide transparent emulation and caching into a single DIMM.

That alternative approach might work better for smaller scale embedded systems which don't have a lot of DIMMs - but creates difficult design constraints - because the "all in a single DIMM" approach means there will be less flexibility about RAM flash cache ratios.

controllernomics and user risk reward with big memory "flash as RAM"

"...Application-unaware design of memory controllers, and in particular memory scheduling algorithms, leads to uncontrolled interference of applications in the memory system"

Are you ready to rethink RAM?