Premise
These six projections are my personal picks of important predictions specifically for 2017. They are all concerning disruptive technologies in the area of infrastructure. The core premise that underlies five of the six projections is the importance of consumer acceptance and volume as a key predictor for successful adoption by the enterprise. It is much easier and far less risky to take a consumer technology and upgrade it for the enterprise than to develop an enterprise technology and ecosystem from scratch.
One key consumer trend is the the slow ending of the PC era as the mobile era ascends. PCs are declining, and with it the volume of Intel x86 chips. An underlying premise is that application development has switched almost completely from x86 to ARM chips driving smartphone, tablets and a myriad other devices. In the near future, no child or teen is going to use a mobile system that does not run the same applications available on iOS or Android.
The reduction in volume from PC x86 affects the cost and investment equations for x86 servers. ARM servers have consumer mobile volumes behind them, in the same way that x86 servers had PC volumes behind them in the 1990s, and Wikibon projects ARM will take 20% of the lower part of x86 over the next five years. Prediction #2 is that 2017 will see the first volume ARM servers.
The bundling of technologies in x86 processors worked well for Intel when consumer volumes and technology turns could allow Intel to achieve higher performance and lower costs across both consumer and server products. With a post-PC era in sight, the lack of consumer volume means other chip architectures will emerge to compete with x86. ARM servers have already been mentioned, and prediction #3 says that ARM processors inside the flash controller will be announced in 2017 to specifically assist map-reduce and other big data applications.
The announcement of OpenPOWER CAPI support for consumer-driven Nvidia GPUs and open source microcode is a game-changer for high-end big data and AI processing. Prediction #4 is that in 2017 Open Power with CAPI and Nvidia GPUs and/or other FPGAs will radically outperform x86 for high-end big data and AI computing.
Traditional slow magnetic disks in consumer systems are being rapidly replaced by another disruptive technology, flash storage, both in PCs and other consumer products. Prediction #5 is that HDD magnetic disk volumes will significantly decline again in 2017.
Prediction #6 is that 3D XPoint will stall in 2017. Wikibon takes this position because there is no announced or even rumored volume use of this technology in the consumer space. The technology was over hyped, and has come in way under the original specifications. It has been slow to achieve any enterprise trials or adoption. Wikibon projects that no high-volume consumer deployment of 3D XPoint technologies will happen any time soon. The consumer market is driving flash and DRAM cost and functionality improvements, and they will be deployed as alternative and cheaper solutions than 3D Xpoint. The likelihood is that 3D XPoint volumes will NOT grow sufficiently to fund lower prices, higher volumes and future investments.
This leaves the only prediction with a cloud theme rather than a consumer theme. Prediction #1 is that AWS will introduce Server SAN Software Storage System for Enterprises in 2017. The reason for this requirement is simple; the cost of moving data, especially large amounts of IoT data, is far more expensive than processing the data locally, as shown in Wikibon’s earlier research. Future systems will need to interoperate in microseconds. The solution is a hybrid cloud, and the key to success is to be able to manage the remote data and processes centrally. Having the same storage software and APIs in both the central and local clouds is a giant step towards cost-effective and performant hybrid clouds.
1/ AWS will introduce Server SAN Software Storage System for Enterprises in 2017
Wikibon predicts that AWS will introduce a software storage system which runs on enterprise provided servers, both on-premises and/or at a colocation facility, with a full set of APIs to manage the storage from AWS. This will initiate the beginning of AWS hybrid systems running both on-premises and in the AWS cloud. This will be particularly important for hybrid Industrial IoT systems which will need to process data locally but manage data remotely.
Next generation datacenter equipment is much smaller and denser, with short high-speed interconnection between processors, memory, GPUs/FPGAs and flash storage. The advantages and projections for a “Server SAN” model has been described in detail in extensive Wikibon research. Edge applications will increasing move from a “system to human operator” model to a “system to system” model, and intersystem response time will move from seconds to microseconds. The speed of light and telecommunication costs and reliability dictate that processing should take place as close as possible to the origin of the data.
One key to successful implementation of hybrid cloud systems is being able to manage the remote site completely from the central site, so that remote IT operational staff are not required. Having the same storage software and APIs in both the central and local clouds is a significant simplification of the distributed environment. Owning the data at both ends in a significant step towards cost-effective and performant hybrid clouds, and makes compliance and performance much easier to monitor and remediate.
2/ Introduction of Volume ARM-based Servers in 2017
2017 will see the first serious announcement and deployment of ARM-based server processors, on the way to owning 20% of the x86 world-wide server revenue by early in the next decade. The biggest benefits of ARM-based servers are the consumer benefits of lower cost, lower power requirements, easier integration with consumer vector systems (e.g., Nvidia), easier integration with sensors (very strong ARM-sensor ecosystem, derived from MEMS sensors in mobile devices) and easier integration with flash storage solutions.
The main constraints on ARM adoption are the strength of the general software ecosystem supporting the x86 architecture and x86-based operating systems. Expect early adoption in specialist systems with very dense compute requirements, in lower-cost supercomputers and in IoT Edge systems.
3/ Big Data and Map-reduce Functionality inside Flash Drives Announced in 2017
2017 will to see the first big data and map-reduce functionality on ARM chips inside flash storage modules. This will dramatically shorten the time and cost to process major long-running big data applications. Rewrite time for Google and Yahoo?
4/ Open Power with CAPI and Nvidia GPUs will Radically Outperform x86 for Big Data and AI computing in 2017
The OpenPOWER Foundation offers a far more flexible architecture for the high-end processing needs of big data, analytics and artificial intelligence. Wikibon projects the announcement of CAPI on OpenPOWER together with consumer-driven Nvidia GPUs and open source microcode will take another 20% of the top of the x86 market over the next five years. The x86 software ecosystem is strong and will provide friction to OpenPOWER general adoption. Intel have completed a major acquisition of Altera, a specialist in FPGAs, to help them in speeding up processing. In August they acquired Nervana, and have been making software partnerships. But Intel are playing catchup with NVIDIA who have been developing for the gaming and mobile marketplaces, and have a large lead. In the high-end computing space with high application value, the top 20% of processor revenue spend, Wikibon projects that OpenPOWER with Nividia will win a high percentage before this decade is over.
5/ Steep Decline of HDD Drives Shipped in 2017
2017 will see the number of hard disk drives shipped decline to below 350 million from 471 million in 2013, a 10% CAGR decline overall. This will be driven by the rapid adoption of flash drives in all consumer devices, led by PC Notebooks with about 50% shipping with Flash-only Drives in 2016. 2016 saw the capacity benefit of HDD drives disappear – the capacity of SSDs reached 15 terabytes for a 2.5” drive, compared with a maximum of 10 terabytes for HDD 3.5” drives. That is the equivalent of a cubic density difference of 5.5 times in favor of flash. Flash is now denser, faster, more power efficient, more reliable, has greater mobility, has longer battery life, more form factors and takes less effort to program and maintain. Flash is more expensive per gigabyte, but consumers increasingly prefer less storage and greater functionality.
This is important to enterprise data centers, because of the long-term negative impact on investment and future functionality of disk drives. Even though cloud providers are buying more magnetic drives at the moment, the continuous reduction in consumer demand will mean a faster adoption of flash to replace HDD drives in the enterprise.
6/ 3D XPoint will stall in 2017
There is no announced or rumored volume use of 3D Point technology in the consumer space. The technology initially over promised, and has been very slow with enterprise adoption. Trial products are slated for 2017, with low volumes. The alternatives of high-speed flash and more DRAM are becoming cheaper by the day, and Samsung and Western Digital are not going to be shy about aggressively pricing alternative DRAM/NAND flash alternatives.
Wikibon projects that unless consumer volume uses are found for 3D XPoint, shipments will stall in 2017. NAND flash and DRAM, together with ARM processors, will continue to advance in performance and price performance. Alternative and lower cost products with DRAM/Flash hybrid memory, high-performance flash and ARM processors, and use of consumer GPUs from Nvidia will make it very difficult for 3D XPoint volumes to grow sufficiently to fund future investment in 3D XPoint.
Action Item:
Wikibon recommends senior IT executives from enterprises and vendors should take into account current consumer acceptance and volumes as key factors in selecting and investing in future enterprise infrastructure technologies.
