Sergey Nivens - Fotolia

News Stay informed about the latest enterprise technology news and product updates.

AIOps early adopters tackle data quality issues

AIOps is new in many ways, but early adopters must also heed a time-honored IT maxim: garbage in, garbage out.

AIOps can augment enterprise IT ops teams as they cope with ever-larger numbers of increasingly complex IT infrastructure components. But AIOps tools are only as good as the data they're given.

The earliest days of AIOps stoked fear that advanced data analytics algorithms attached to automated machines will replace human IT experts, but those fears remain far-fetched at best. Early adopters say AIOps tools are far from a magic bullet, and IT ops jobs are safe, even as organizations use artificial intelligence and machine learning tools to sort through infrastructure monitoring data, reduce alert noise and, in some cases, investigate or resolve the causes of incidents. The effectiveness of AIOps software also remains limited by how solidly human IT pros build the data pipelines that feed it and how well human operators in IT and business interpret its results.

"In many situations, we help customers realize they don't actually have the right data in place," said Amer Deeba, COO of Moogsoft, an AIOps software vendor in San Francisco. "It's not just the alerts that are coming in, but the enrichment of the data, how you can connect data as it's coming in to create context."

Enterprises realign data repositories, pipelines for AIOps

Before KeyBank put Moogsoft's software into production in late 2017, it had to rethink the IT monitoring tools it already used.

"Two years ago, I walked into a very typical IT company, and monitoring was like Noah's Ark -- we had two of everything," said Mick Miller, senior DevOps architect at the financial services company in Cleveland. IT admins switched between 22 different tools and those tools' respective data repositories. "KeyBank was really good at MTTB -- mean time to blame," he said.

KeyBank whittled its monitoring tools down to Dynatrace to monitor application performance and end-user experience, and Elastic Beats* to monitor the network, servers and logs. It also built a data pipeline with the Apache Kafka distributed messaging system and Elasticsearch indexed data repository that partially filters raw monitoring data before it's passed on to Moogsoft. The pipeline data enrichment tier also arranges clean data into a standard schema, which is critical to effective data analytics, Miller said.

Dynatrace touts its own AIOps features, including automated root-cause analysis, but KeyBank doesn't use those features, in part, because its per-agent pricing would make comprehensive monitoring for the entire KeyBank environment prohibitively expensive, Miller said. KeyBank also considered Moogsoft AIOps competitors ServiceNow, IBM Watson and BigPanda, but found Moogsoft's data science expertise and algorithms most impressive two years ago, he said.

OpsRamp Observed Mode
OpsRamp's summer 2019 update includes Observed Mode, which can perform 'dry runs' of AIOps tasks to test their effectiveness.

AIOps tools beef up data visualization features

Other AIOps early adopters reported similar experiences with data quality revamps before AIOps tools delivered on their promise.

We used to set and forget most monitoring and didn't touch the tool configuration unless something broke, but [AIOps] requires more tuning and better quality data.
Tim HebertChief managed services officer, Carousel Industries

Carousel Industries, an IT managed service provider based in Exeter, R.I., has used AIOps software from OpsRamp for two years and had to integrate OpsRamp with its ServiceNow change management database through APIs to get effective results.

"We learned that the number of times people had to touch the data and reenter it between systems, the more errors there were," said Tim Hebert, chief managed services officer at Carousel. "We also used to set and forget most monitoring and didn't touch the tool configuration unless something broke, but [AIOps] requires more tuning and better quality data."

Carousel's service delivery team manually reviews 1% of IT service desk tickets automatically processed with OpsRamp daily to ensure their accuracy and consistency. Once a quarter, Carousel IT personnel meet with OpsRamp to offer suggestions for new features. In the past, such suggestions yielded deeper telemetry collection on Avaya unified communications devices, among other updates.

Carousel has waited for OpsRamp's summer 2019 release, rolled out this month, for service and topology maps and observed mode features that clearly visualize the relationships between applications and IT services and perform "dry runs" of AIOps algorithms to ensure they behave as expected. Hebert said he hopes these features will further reduce the manual intervention required of his company's IT admins.

"There's still about 50% of our repetitive manual tasks that we can eliminate," Hebert said. "For some of these events, we could automate the creation of service ticket. But because of the complexity or a lack of information about the system, a human decision was needed to resolve them."

* Information updated after publication

Moogsoft also rolled out data and workflow visualization tools with version 7.2 of its AIOps platform in late May 2019.

"We get some nice visualizations and see how the AI data collection and our workflows around it are working," KeyBank's Miller said of the new release. The company plans to put that release through user acceptance testing this month, he said.

Dig Deeper on Real-Time Performance Monitoring and Management

Join the conversation

1 comment

Send me notifications when other members comment.

Please create a username to comment.

How have you ensured data quality for AIOps in your organization?
Cancel

-ADS BY GOOGLE

SearchDataCenter

SearchAWS

SearchServerVirtualization

SearchCloudApplications

TheServerSide.com

SearchCloudComputing

DevOpsAgenda

Close