By Charles King, Pund-IT, Inc. June 20, 2018
Hardware/software/data integration and interdependencies are important for enterprise workloads but are especially critical when it comes to performance-sensitive applications, such as artificial intelligence (AI). Unfortunately, they’re also easy points to misunderstand when highly complex technologies are in the early stages of commercial development and deployment like, again, AI.
As a result, if or when organizations move to or beyond AI proof of concept (PoC) exercises without clearly knowing the challenges and risks they face, it’s all too easy for them to run into problems, fail unnecessarily, then scale-back or abandon their efforts. So, it’s great when vendors help customers anticipate and steer clear of avoidable pitfalls with solutions designed to contend with and overcome fundamental technological complexities.
Those points came to mind regarding new offerings from the IBM Cognitive Systems and IBM Analytics groups. Let’s consider those announcements separately, along with how they’ll affect the company’s analytics, AI and other offerings, and customers’ related efforts.
IBM Power Systems – A reference architecture for AI
Despite the considerable hype being directed at AI’s commercial possibilities, the vast majority of businesses are still in very early stages of exploring the technology and its potential impact on their businesses. There are multiple reasons for this but prime among them is the complexity of most solutions, including hardware/software stacks and workflow/data flow processes.
As a result, pursuing and succeeding in AI requires technological sophistication and “roll your own” IT skills that are beyond the capabilities of many, if not most companies. To help address those issues, IBM’s Cognitive Systems group introduced the first iteration (v 1.1) of PowerAI Enterprise and a related reference architecture for on-premises AI deployments.
What exactly is PowerAI Enterprise? It’s a platform that adds five new development tools to IBM’s PowerAI distribution of machine learning (ML) and deep learning (DL) frameworks. They include,
- A structured, template-based approach for building and transforming data sets that shortens transformation time
- Data import tools, including an integrated resource manager that can optimize jobs for low cost (fewest nodes/cores) or fast execution (more nodes/cores)
- An integrated quality step function that quickly tests the clarity of data signals and detect obvious issues or deficiencies in training data sets
- Model set-up tools designed to select the most promising combinations of hyperparameters, thus reducing non-productive early runs
- The ability to visualize the progress of training jobs and warn data scientists and model developers when things go wrong
According to IBM, taken together these new functions and capabilities can reduce wasted time, effort and investments, accelerate project development and increase the overall efficiency of data scientists.
How about hardware support? That’s where IBM’s AI Infrastructure Reference Architecture comes in, an integrated platform based on the company’s Power Systems AC922 (for compute) and LC921/LC922 (for storage) accelerated servers, software and tools. Together, they provide optimal support for AI-related data preparation, model training and inference processes.
IBM emphasized that developing these new offerings resulted from the company’s engagements with hundreds of customers, ranging from PoC projects to fully implemented production services. That includes Wells Fargo which uses IBM solutions to perform deep learning models to comply with critical financial validation processes.
Given the $1B in fines recently levied against Well Fargo for past customer abuses, improving financial compliance and oversight is obviously top of mind. IBM’s PowerAI Enterprise is the kind of solution that should help the company achieve those goals.
IBM Analytics extends Hortonworks relationship to the cloud
At first glance, the news from IBM and Hortonworks may not appear specifically related to AI. In a blog, Rob Thomas, GM of IBM Analytics, noted that the pair are deepening their relationship (announced a year ago) melding of the Hortonworks Data Platform (HDP) and IBM’s Data Science Experience.
The goals of that effort were to expand the data science capabilities and features of Apache Hadoop file systems, including data lake environments while extending the use of HDP among enterprises. Along with providing HDP-based offerings and services for IBM customers, the solution is also part of the portfolio offered by Truata, a European analytics trust company launched in March by MasterCard and IBM that assists companies in meeting the requirements of the EU’s recently-implemented GDPR.
Thomas’ stated that following the past year’s successes the pair are introducing IBM Hosted Analytics with Hortonworks (IHAH), an offering now available via IBM Cloud that integrates HDP with IBM’s Data Science Experience and its Big SQL data warehouse system for Hadoop. IHAH provides customers a fully provisioned, cloud-enabled environment for data management and analytics, simplifying and speeding setup, provisioning, security and deployment processes for Hadoop-based data analytics.
That’s all fine and good but what does it have to do with AI? To date, many if not most AI-focused public announcements, discussions and marketing efforts have emphasized hardware-oriented features and capabilities. That makes a certain kind of sense since until relatively recently, AI was simply too complex, costly and time-consuming to be commercially workable or sustainable.
Hardware-related advances, including the system-level integration of CPU and GPU technologies featured in IBM’s Power Systems AC922 servers and other solutions have finally made AI cost-effective and achievable. But at the same time, unified data preparation, management and governance, robust analytics capabilities and innovative information architecture technologies are essential to the success of AI-related ML and DL processes.
Providing easy access to integrated, optimized cloud-based data platform solutions for advanced analytics projects, including artificial intelligence, is what IBM and Hortonworks’ new IHAH solutions are designed to deliver.
Not so long ago, the concept of “system vendors” held real weight in IT and other industries with vendors delivering real value to their customers via optimally performing data center solutions tuned for customers’ specific needs and requirements. That perception changed fundamentally with the increasing commoditization of server and storage components and the rise of public and private cloud computing.
However, the emergence and rapid evolution of advanced analytics technologies and AI solutions and services may well result in a reconsideration of the “systems” value vendors offer their customers. Some of that will certainly be hardware-centric with server and storage solutions designed and tuned for machine and deep learning processes.
But a larger opportunity exists for vendors that have the expertise, assets and willingness to further invest in the software and tools essential maximize the value of their hardware solutions and ensure the success of AI. That includes systems that reside on premises in enterprise data centers and in the cloud-based platforms enterprises use to support their projects.
These new solutions, services and partner-enabled offerings from IBM Cognitive Systems and IBM Analytics demonstrate what willing, experienced vendors can accomplish. Where the company and its customers go from here deserves close attention.
© 2018 Pund-IT, Inc. All rights reserved.