IBM Research pioneers technologies behind new AI for IT capabilities
As Information Technology (IT) complexity grows and the use of AI technologies expand, enterprises are looking to bring in the power of AI to transform how they develop, deploy and operate their IT. IBM is launching today a broad range of new AI-powered capabilities and services to help CIOs automate various aspects of IT development, infrastructure and operations, including:
- IBM Watson AIOps: leverages AI to reliably operate enterprise applications and automate the detection, diagnosis and response to IT anomalies in real time.
- Accelerator for Application Modernization with AI: a suite of tools within the Cloud Modernization Service designed to reduce the overall effort and costs associated with application modernization through advanced AI technology from IBM Research. This accelerator leverages continuous learning and interpretable AI models to adapt to a client’s preferred software engineering practices and stays up-to-date with the evolution of technology and platforms.
As is the case with much of IBM’s AI development, significant portions of the technologies underlying Watson AIOps and the Accelerator were born out of IBM Research. This new offering and service — part of what we’re calling AI for IT — is the culmination of years of research and development at IBM Research into how AI can be used to transform the IT lifecycle.1, 2, 3, 4, 5, 6, 7, 8
CIOs and their application Site Reliability Engineering (SRE) teams are overwhelmed by the sheer number of tools for operations with data fragmentation across them and the complexity of issues, making IT operations a challenging domain. Watson AIOps reimagines IT operations with AI by enabling the prediction of problems and offering the potential of proactively fixing them. Breakthrough AI techniques1, 2, 3, 4, 5 developed by IBM Research enable Watson AIOps to discover new patterns in IT operations, remove noise, correlate problems across multiple data sources and make recommendations to fix them.
More specifically, the innovative tools and techniques in Watson AIOps correlate various multi-modal signals in IT operations by leveraging structured, semi-structured and unstructured data sources for building correlation models. For example, when an enterprise application experiences a problem, many monitoring tools start emitting a swarm of alerts, sometimes numbering in the thousands. Often, the root cause of these incidents lies in a different spot than the alerts. In fact, many times the swarm of these alerts distract SRE teams, wasting precious time during which the application continues to suffer operational instability or, even worse, failure.
SRE teams typically evaluate a multitude of data sources including metrics signals, application and system logs, the swarm of alert signals, and even past incident tickets. Many existing tools in IT operations can individually build models in each of these data silos to predict continuous values of the input signals or predict probabilistic class labels. But, they still treat the input spaces in isolation, resulting in hundreds of anomalies and, again, overwhelming the support teams. This is precisely where AI technology behind Watson AIOps steps in. It correlates among the diverse data sources to localize the real root cause, create an explainable diagnosis and recommend the best course of action.
To do the above correlation, the algorithms need to work with time series data of metrics, semi-structured — but voluminous— data logs, structured data like alerts, and unstructured data in incidents and human conversations to automatically create a timeline of the evolving issue. Each of these data sources better lends itself to certain types of tasks. Time series data, for example, is more suitable for regression tasks, whereas unstructured data is best for classification tasks. Logs and other semi-structured data can be used for either of the tasks after suitable transformations. For SRE teams to resolve the incident impacting the enterprise application, they must first uncover patterns across these diverse, multi-modal signals in order to reduce the problem from thousands of alerts to hundreds of anomalies coalesced into a few possible evolving stories.
Watson AIOps also leverages semantic search techniques that can relate the current incident to past incidents, analyzing those contextual cases, and suggesting possible next best remediations. IBM AI innovations, like Watson OpenScale, are at the forefront of developing trusted and explainable technologies, and we leveraged those innovations tohelp SREs interpret the reason behind a Watson AIOps recommendation which is critical to trusting those actions.
A major hurdle CIOs face in leveraging the cloud paradigm with core business applications is that these applications were written for an on-premise world and not architected to leverage cloud-native architecture and modern DevSecOps principles. Application modernization is about optimizing an enterprise’s application portfolio and transforming it to meet the rapidly evolving needs of business agility and competitiveness, all while leveraging these new programming models and cloud architecture.
Many large enterprises have thousands of legacy application that must be moved to the cloud. As enterprises move these mission-critical workloads, they face difficult decisions about containerization. CIO teams making these decisions must consider, for example, application criticality, behavior, operational requirements and hosting infrastructure. This decision-making process is largely manual and often error-prone when done at scale.
The Accelerator suite consists of three tools in which IBM Research led development:
Application Containerization Advisor (ACA): This asset uses AI to quickly provide more confident recommendations for containerization. ACA applies extensive and evolving knowledge graphs to help infer missing data and compute matches to the possible container references. The AI-enabled advisor also employs continuous learning models to provide highly accurate containerization recommendations that improve over time. The tool also considers 12-factor properties to provide insights into the complexity of containerization activity, and its AI Explainability features help IT departments understand assessments of feasibility, complexity and risk.
Candidate Microservices Advisor (CMA): Legacy applications typically bundle functionality across multiple business and data domains into a single deployable application, i.e., they are implemented as a monolith. This severely restricts a business’s ability to roll out frequent feature releases that may impact only some pieces of functionality or domains. It is also difficult to scale the performance of certain functionalities up and down selectively and dynamically, as needed by changing usage load levels. CMA automates the discovery and design of candidate microservices from the source code and data artifacts of a monolithic application. It also captures application details including component entities, for example, Java classes and database tables, as well as the relationships between these entities and the transactions contained within the application. The idea is to determine whether and how the application can be broken down into microservices that can be moved to a hybrid cloud environment.
Modernization Workflow Orchestrator (MWO): This functions as an AI-driven system to introduce application modernization at scale through standardized tools and architectural best practices. Think of Modernization Workflow Orchestrator (MWO) as a GPS for modernization. MWO employs AI planning technology based on symbolic reasoning to dynamically create modernization steps tailored to each application. The tool also leverages Natural Language Processing (NLP) and Machine Learning (ML) to accelerate the capture of knowledge about modernization actions for the AI planner.
The availability of Watson AIOps and Accelerator for Application Modernization with AI is another example of IBM Research helping the company move quickly from innovation to product. IBM Research has also helped develop many of the NLP capabilities driving Watson Discovery for document understanding and Watson Assistant for virtual agents. This includes our announcement in March that IBM will begin integrating NLP features from IBM’s Project Debater into Watson, ultimately enabling organizations using Watson Discovery, Watson Assistant and Watson Core Services to take advantage of advanced sentiment analysis, new summarization capabilities, advanced topic clustering and customizable classification of elements in business documents.
IBM Research will continue developing ways to leverage AI to transform information technology, to accelerate the process of understanding what an organization’s application portfolio looks like and help them decide how to modernize it and deploy in the cloud. The new era of AI for IT is here.
Date
References
-
Automated Dispatch of Helpdesk Email Tickets: Pushing the Limits with AI. Mandal et al., AAAI 2019. ↩ ↩2
-
MTEX-CNN: Multivariate Time Series EXplanations for Predictions with Convolutional Neural Networks, Assaf et al, International Conference on Data Mining 2019. ↩ ↩2
-
Online Interactive Collaborative Filtering Using Multi-Armed Bandit with Dependent Arms. Wang et al, IEEE Transactions on Knowledge Data Engineering, August, 2019. ↩ ↩2
-
An Integrated Framework for Mining Temporal Logs from Fluctuating Events. Zeng, et al, IEEE Transactions on Services Computing, February 2019. ↩ ↩2
-
Transforming the IT Services Lifecycle with AI Technologies. , Kloeckner et al, Springer 2018, ISBN 978-3-319-94047-2, pp. 1-100. ↩ ↩2
-
Method and System to AI-Infused Transformation of Legacy Applications toward Modernization, Kalia et al. , US Patent 16/823649. ↩
-
Generation of microservices from a monolithic application based on runtime traces, Xiao et al. US Patent 16/855565. ↩
-
Towards Automated Planning for Enterprise Services: Opportunities and Challenges, Vukovic et al. International Conference on Service Oriented Computing 2019. ↩