Data Center Journal

VOLUME 57 | OCTOBER 2018

Issue link: https://cp.revolio.com/i/1036193

Contents of this Issue

Navigation

Page 6 of 20

4 | THE DATA CENTER JOURNAL www.datacenterjournal.com connectivity, for example. Each side may therefore govern those aspects of data center design most in line with its expertise. As companies consolidate their data centers, however, real-estate providers stand to benefit from more than just cloud companies. Cloud providers in many cases can deliver an IT service with greater efficiency than a customer for which IT is a periph- eral duty. But at some large scale, the efficiency balance may tip back toward customer-owned or customer- provided service. At that scale, the best option for some companies is to bypass public-cloud providers and go directly to real-estate providers. Some competition thus emerges as cloud and real-estate providers design to meet the needs of such customers. CONSOLIDATION AND LATENCY A growing concern as data center consolidation persists is latency. Imagine the extreme case: all data center resources reside in a single geographical location. e problem is that service delivery is limited by the time that signals require to travel back and forth from the customer to the provider (i.e., the speed of light). In some cases, long latencies are simply annoying; in others, they're financially costly. e buzzword in response to this concern is the edge: the need to keep resources close to the customer to reduce latency. e result is a force that coun- teracts geographical consolidation, but owning and operating multiple (even small edge-based) data centers, particularly in several countries, is difficult. Large companies, whether a cloud-service provider such as Amazon or a real-estate provider such as Digital Realty, ostensibly have more resources to "line the edge" with data centers. But a shotgun distribution of edge data centers may be a better fit for a real-estate provider than for a cloud provider, depending on the pro- vider's scale, market, capital and so on. In either case, however, the problem of edge computing will be difficult to tackle and calls for design strategies beyond simple consolidation. DESIGN AT THE IT LEVEL Although silicon-technology innovation continues, the industry has fallen behind the pace of Moore's Law. e implication is that increasing IT resources at the same rate as before now requires more than just waiting for the next process technology from Intel, Samsung, TSMC or another foundry. us, innovative design at the IT level is more important than ever if service providers—whether cloud or otherwise—are to keep up with demand. In this regard, large cloud-service providers are in an excellent position to try new approaches. For example, whereas CPUs were once the primary or even exclusive compute unit, GPUs, neural-network accelerators and other specialized processing units are taking on a growing role. GPUs, for instance, excel at parallel processing and increasingly serve in supercomputers where so-called scale-out workloads are common. Also, neural-network accelerators help support the many artificial-intelligence applications that are burgeoning in various industries. Search giant Google is one example: it developed its "tensor processing unit" (TPU) to accelerate machine- learning applications. Such designs aren't replacements for CPUs, but they illustrate the proliferation of specialized architectures built to tackle certain problems more efficiently than standard architectures. e end of Moore's Law is therefore an exciting development in the sense that it may spur greater innovation rather than reinforcing the standard brute-force approach to computing (i.e., just shrink the transistors). is aspect of design falls more in the purview of cloud providers than data center real-estate providers, as it's closer to the heart of cloud services as such, but it illustrates that the two sides can work in complementary fashion. e future of data center design is therefore in the hands of both cloud and real-estate companies, at least as long as the trend toward consolidation continues. CONCLUSIONS Data center consolidation, driven in large part by a growing move to cloud computing, places major design decisions into the hands of cloud-ser- vice providers as well as the real-estate providers that house many of them. e economics of cloud computing suggest this trend will continue, as does the end of Moore's Law. Compa- nies for which IT is a peripheral duty usually lack the resources and knowl- edge, or both, to develop innovative computing technologies and data center designs. ese tasks therefore fall to cloud and real-estate provid- ers, which are generally in a better position to tackle them. Between these two groups, however, innova- tion is likely to be complementary rather than competitive, with each one applying its strengths to address the growing demand for IT resources despite flat IT spending. n

Articles in this issue

Links on this page

Archives of this issue

view archives of Data Center Journal - VOLUME 57 | OCTOBER 2018