Flash technology bottlenecks still difficult to overcome is doomed to just passing through?

Fujitsu Chief Technology officer Dr Joseph Reger

Flash memory is just a central station for our memory storage architecture--a point that Dr. Joseph Reger, Fujitsu's chief technology officer, recently raised.

Flash is facing some unresolved problems, according to Dr. Joseph Reger, chief technology officer of Fujitsu's technology solutions. He says the density of flash memory is increasing, but this is in exchange for the ability to read and write data. With the improvement of manufacturing process, the durability of flash is declining inversely, for example, when the MLC of the number of multilayer unit flash from 2-bit to 3-bit to 4-bit, its durability is 1/2 to 1/3 to 1/4 of the decline, they are facing the contradiction between speed and durability.

Flash--Destined to be just a traveler?

Even before that, ebay has replaced about 100T of mechanical hard drives in traditional storage arrays as SSD solid-state drives, with significant impact--a 50% reduction in storage rack space, reduced power consumption, 5 times-fold 78%,i/o performance, and a 5-minute deployment of a new virtual machine. In the past it took 45 minutes.

But it's not enough in Reger's eyes.

"You can make the control chip smarter, like the SandForce company," he said. But this also does not solve the problem fundamentally, but the surface solves the problem, the flash memory is not the end point but only an intermediate stage. "As time goes on, flash technology will be history," Reger said. We need to explore another set of new technologies to get a new tier of storage in a storage hierarchy.

In Reger's view, the ideal storage hierarchy should be--when crossing from one type to another--the speed of access should be an order of magnitude jump. Over time, PCM (phase change storage phase changing Memory) is the technology that best represents next-generation storage trends than other competitors, such as HP's Memristor technology.

SandForce SSD Control chip

It is worth noting that if the flash is eventually replaced by memory and other storage technologies, such as SandForce and Anobit, the technology start-ups focusing on flash storage technology will be in trouble. Their technology solves some of the current problems with flash storage, but their technology will no longer be valuable if flash technology is eliminated.

In addition, while all the flash arrays now represent state-of-the-art optimization techniques and thrive for some time, there is a massive sharing problem today. Flash storage is not the best choice for these applications if we are faced with a large scale of SAN and NAS data warehouses.

Is PCM the future?

In 1960, Dr. Stanford Ovshinsky, an American inventor, found that some glass changes in the phase change resistance, which is reversible; several years later, he found that some of the materials in different phases, the reflectivity of the laser is also different. These discoveries mean that people can measure the phase of the material by current or laser, and that there is a new way to develop storage devices--phase change memory.

Now the universal design of the phase change memory is to clamp a layer of GST between the top electrode and the bottom electrode, and the heating resistor extending from the bottom electrode to contact the GST layer. Current injection heating resistance and GST connection point, the resulting heat will cause phase change, the material properties of the phase change by the current, voltage and time, can be written with a strong current, read with a weaker current.

This storage device has many advantages. It has a lifetime of 10 million write cycles, much higher than the enterprise-class flash chip 30,000 times, it can store the smallest unit is 1 bits, which is the smallest unit of data that people use to measure. Unlike memory, which requires a constant supply of current to not lose data, the speed of reading and writing is much higher than that of flash memory, while bandwidth can be comparable to RAM. In any case, it's like the kind of storage device people want.

But there is still a problem that has plagued the development of the phase-change memory: its storage unit can only be stored in one place, the cost is not low and the capacity is not high, currently only small scale for mobile phones, not suitable for computers.

Full Flash Array

NAND process and time comparison

Take a look at the current flash memory, such as Nimbus, violin and Huawei Symantec, and other manufacturers full flash array products are designed to serve disk-intensive I/O applications. The flash control chip is optimized by software. Of course, the hard disk storage controller will also do software optimization. However, these two code heaps will result in a full flash array and a hard drive array management problem. This means, for example, the need for some kind of abstraction layer on top of the flash array and the hard disk array, where each type of storage detail is hidden on the stack.

Massive Solid state data storage

In the future of Flash, our view is that the advantages of a full flash array will lead to a conflict between the operation and management of the flash array and the operation and management of the disk array. The reduction of each process and the increased number of cell levels will lead to new problems with flash durability, speed, and error handling. But this benefits the future development of a large flash array, large enough to hold the entire application and its data in memory.

For example, there is a potential hard drive structure in the array, which creates a new paradox because there is no requirement to write data and read data. And because the data is not important, do you want to copy the content to disk? Perhaps the answer is yes, because flash and solid-state storage will replace online storage, and disk as a near-line storage and offline repository. The cost advantage of the disk is unmatched.

The large Flash array opens the door to the memory architecture, which will firm the future with a technology that replaces flash memory, either. Without a permanent technology, flash technology lays the foundation for solid-state storage and non-volatile technology, and in the future it is possible to radically change the way we handle, manage, and protect large amounts of data.

The future is still mixed storage media

Reger is focusing on an upcoming new storage architecture and similar database. How to deal with data access has become our challenge. There is no need to use technical ideas based on disk access, such as paging, where the memory database may be smarter.

In the near future, we can simply trust the memory we put into the server, because everything can be in memory? Imagine what this means for memory management and storage management when you have 2TB of memory?

If everything is rewritten and redesigned with data storage management work. There will be only one effective tier of storage? or some form of memory?

At present, data storage does not mean that data is stored in memory, will there be any change in the future? After all, storage was invented to overcome the limitations of memory size. If the size of the memory limit is broken then will anyone need to store it?

Reger believes that the future system architecture will be data oriented rather than memory and storage, but for now it is still unreachable. In the next few years, he argues, flash memory does not completely replace traditional storage in the datacenter. But it could be a big percentage. and used in certain important environments. For example, it requires high iops (Input/output Operations per Second) or a place that requires efficient energy.

He reckons that the energy efficiency of flash memory is 3 orders of magnitude higher than the traditional hard drive, which means that there are 2 orders of magnitude higher in ioPS watts and more PCM. But even so, the traditional hard drive will not go away for the next 10 years.

(Responsible editor: admin)