Virtualization Technology News and Information
Turbonomic 2017 Predictions: IT Industry Predictions from the Experts

VMblog Predictions 2017

Virtualization and Cloud executives share their predictions for 2017.  Read them in this 9th annual series exclusive.

Contributed by Yuri Rabover, Yechiem Yemini and Eric Wright, Turbonomic

IT Industry Predictions from Turbonomic Experts

Once again, we asked Turbonomic co-founders Yechiam Yemini (YY) and Yuri Rabover, and our Principal Solutions Engineer and Technology Evangelist Eric Wright, aka @discoposse, for their predictions. 

Yechiem Yemini

Big Data Systems Will Need to Integrate With IT Operations

Big-Data AI applications, will increasingly populate enterprise IT and mature from testbed studies into business critical  systems.  IT operations management will need Big-Data extensions to:  

(a) assure the performance of Big-Data applications; and

(b) support their cost-efficient scalability.

Such extensions, of traditional operations management to handle Big Data, are far from trivial. Traditional IT partitions resource management between OS schedulers, controlling real-time resource allocations to workload demands, and operations management systems, managing static resource configurations and resource performance. Applications can exercise indirect control by invoking OS and/or operations management API. This control is necessarily limited because of the hazards of potential interference by resource sharing "neighbors."

In contrast Big Data applications depend on finely tuned resource visibility and control. Consider a small number of examples:

  1. Big-Data tasks must be placed, possibly dynamically, within low-latency access to their data. Even a minor change in access latency may accumulate into hours delays.  Location control requires visibility of topology, latency metrics and data access metrics. Some of these parameters may change dynamically requiring dynamic control mechanisms.

  2. Big Data systems often split workloads among parallel tasks, running at containers; the resources allocated to these containers must be tuned to handle their dynamic workloads demands. Such dynamic resource control requires extensive resource monitoring as well as dynamic real time allocation to match supply and demand for resources.

  3. These parallel containerized tasks , e.g., Mappers of Map Reduce application, are often required to synchronize their schedules. For example Mappers must wait for the slowest ones (stragglers) to complete, before proceeding to the next stage. Stragglers result in delays and in wasted resources held by tasks waiting for them. These stragglers delays and waste can be extremely sensitive to resource scheduling. For example, even minor interference by a noisy neighbor may significantly amplify straggler delays. The resources allocated to containers must thus be carefully scheduled to control the delays and waste by stragglers.

At present, Big-Data systems integrate applications code with ad-hoc operations support. This leads to heterogeneous, complex and costly operations management problems.  As new Big Data applications emerge and introduce new ad-hoc operations management procedures, the complexity and Opex costs can set high barriers for deploying Big Data applications.

Furthermore, in the absence of common, interoperable operations management support, the decision between building private Big-Data clouds or using public clouds, or the decision which public cloud to use may become intractable. Competing standards - e.g., "should we use YARN or Docker" for our Big-Data system?" --render these decisions even messier. 

We predict that 2017 will be a turning point in growing novel technologies to extend IT operations management to handle Big-Data systems.


Yuri Rabover

1.  Datacenter capacity consolidation - The economy of scale kicks in. While the jury is still out how aggressive the move to the public cloud will be, consolidation in the private datacenters has been going on for a while and will be only increasing. Enterprises create larger datacenters and eliminating many regional and smaller ones, as advances in network technologies and available bandwidth reduce the need in local presence. Hardware vendors now offer leased flexible computer capacity right on customer prems or in regional SPs serving multiple customers. One way or the other the economy of public clouds on prem becomes more and more mature and attractive and can serve as a viable alternative to the move to the public cloud.

2.  Datacenter OS spread - Related to the previous trend there is a growing need to better ways of managing pools of consolidated resources. It is no longer feasible to keep partitioning the consolidated capacity per function, customer etc as it grossly impacts efficiency. Solutions like Mesos, K8S and others will become more mature and feature rich offering services not only to the new breed of cloud native apps but also to the legacy traditional apps requiring better access to capacity. Capacity as a service will be becoming more visible and DC/OS offering such service will be playing a more significant role.

3.  Hybrid clouds are here to stay - Hybrid clouds once considered only as a transition period to public clouds will continue playing a visible role. With the consolidated capacity and capacity as a service it will become easier to leverage the existing onprem (owned or leased) resources and use the public cloud based purely on economic but not technology reasons. With partnerships like VMWare-AWS or WMware-IBM the boundary between public and private clouds will become fuzzier and customers can decide practically in real time where and when to use available capacity. It will require additional advances in SDN and SDS which are happening as we speak, technologies like NSX are becoming more mature and feature rich.

4.  IaaS, CaaS, PaaS will continue to blend - The boundaries between these services will be becoming fuzzier and fuzzier and will be more and more driven by the application needs rather than pure infrastructure offerings (being it traditional IaaS or container=based offerings). The latest progress in containerizing Openstack illustrates it well. Which particular brand or service will prevail remains to be seen but the focus will be shifting more and more to the consumers of these services with more and more focus on application development and deployment. What will stay underneath will become less visible and more utility like until eventually becoming an integral functional part of the consolidated offerings

5. The need in autonomous performance control solutions will be getting stronger - With all the trends above using traditional methods of assuring performance not only will become more difficult - it will become absolutely impossible. The scale of the environment and resource service offerings (be it consolidated HCI datacenter capacity or DC/OS on top of it) will make all existing approaches obsolete and force more and more demand to autonomous resource allocation control where consumer of resources will determine how much they need depending on service levels, end user demand etc. Service offerings featuring such capabilities will have a clear advantage over traditional static and reactive ways as eliminating expensive and slow manual labor and greatly increasing efficiency and agility of resource offerings emerging along the lines of previous trends.


Eric Wright

1.  Intel will become the center for much of enterprise IT - Beyond the chip, we are seeing a rise in the software side of Intel.  That includes many interesting projects that are under the open source platforms.  There is a need to keep up the pace as non-Intel chips are were getting some focus as we saw some inklings of ARM-based servers rising in some of the large-scale data center environments.

2.  Open Source at the Enterprise - We have the largest technology vendors creating and contributing open source projects at a pace which is unprecedented compared to previous years.  This momentum will be maintained and increase in 2017 with no end in sight.  The ability to iterate with the community on software is not only valuable, but has become table stakes for many companies who have "prefer open source" in the requirements list.

3.  OpenStack is Not Dead Yet - Despite the many challenges that seem to have the pundits dancing on the grave of the open private cloud platform, I'm long on OpenStack and see that we are going to make big strides as projects like the Intel and CoreOS packaged delivery with Tectonic, plus open projects that bring what Stackanetes can to the fore.  By handling the underlay tooling with a common abstraction, it makes delivering the IaaS portion of the infrastructure solution much easier. Also, watch for what Rob Hirschfeld is doing at Digital Rebar ( for this not just OpenStack, but potentially much more.

4.  DevOps Lite - The journey to DevOps is one that never ends.  It does have a few stops along the way as we see more measurable success with what I term as DevOps Lite.  This is where we have more cross-silo interactions, and more effective tooling that will solve some of the needs of both the Operations and the Development teams at the same time.  Searching for the unicorn-like full DevOps shops will finally stop, and we will begin to make the turn towards embracing the partial shift.  In my mind, being a DevOps Lite is better than no DevOps practices at all.

5.  Security. Security. Security.  Did I mention Security? - This is recycled from last year, but the truth is that we are seeing much more focus already, with a growing push towards better practices, better awareness, and the acceptance that we have to do more.  As a track cyclist, we say "there are two kinds of track cyclists:  those who've crashed, and those who are about to".  In the world of IoT and most of our businesses being internet-facing in one way or another, the new phrase is "there are two types of infrastructure platforms:  those which know they have been breached, and those who are about to be."


About the Authors


Prof. Yemini co-founded Turbonomic, Inc. in 2009 and served as its Chief Scientific Advisor. Prof. Yemini has served as a director and advisory board member of several high-tech companies and venture funds, and as a member of several government technology commissions and working groups.  Prof.  Yechiam Yemini was also a co-founder of Comverse Technology, Inc., Arootz Technologies Ltd (now known as Intercast Networks Ltd), and System Management Arts (SMARTS) acquired by EMC in 2005.  He authored over 200 publications and 15 patents, and lectured widely in these areas. Technologies created at his lab have been widely exported to thousands of sites and commercialized by several companies.


Before joining Turbonomic, Yuri Rabover managed the Advanced Solution group in EMC's CTO office and worked closely with EMC's Architecture and Applied Research teams in prototyping and innovating across a broad range of technology stacks and solutions. Yuri joined EMC with the acquisition of Smarts, where he had a long and diverse career: he served as a member of the founding team, managed engineering and product development, and was responsible for technology partnership development, managing relationships with strategic accounts.


Before joining Turbonomic, Eric Wright served as a systems architect at Raymond James in Toronto. As a result of his work, Eric was named a VMware vExpert and Cisco Champion with a background in virtualization, OpenStack, business continuity, PowerShell scripting and systems automation. He's worked in many industries, including financial services, health services and engineering firms. As the author behind, a technology and virtualization blog, Eric is also a regular contributor to community-driven technology groups such as the vBrownBag community and leading the VMUG organization in Toronto, Canada. He is a Pluralsight Author, the leading provider of online training for tech and creative professionals. Eric's latest course is "Introduction to OpenStack" you can check it out at

Published Tuesday, December 13, 2016 9:03 AM by David Marshall
There are no comments for this post.
To post a comment, you must be a registered user. Registration is free and easy! Sign up now!
<December 2016>