Unravel Data Systems Inc. is updating its application performance management system for big data environments with improved detection of runaway applications, better service level agreement management and enhanced ability to diagnose and recommend fixes for slowdowns and other performance problems.

The company, which emerged from stealth about six months ago with $7.2 million in funding, says it’s tackling a problem that conventional systems monitoring tools can’t. It provides automated problem discovery, root-cause analysis and resolution specifically for big data environments, including elements such as such as Hadoop, Kafka, Spark, Hive and Impala. The company aims to simplify a process that is currently distributed among many different monitoring and administration tools, such as application logs, Cloudera Inc.’s Manager, MapR Technologies Inc.’s Control System and the Apache Sparkcontext.

“This is App Dynamics for the data stack,” said Kunal Agarwal, chief executive and co-founder, referring to the richly funded performance monitoring company that was acquired by Cisco Systems Inc. just before its initial public offering in January.

Unravel Data is targeting an emerging discipline it calls “DataOps,” a term derived from the DevOps agile development technique that gives developers control of both applications and the environment in which they run. Like DevOps, DataOps packages big data jobs and the resources they need. However, data administrators often over-provision to minimize the risk of slowdowns, which creates inefficiencies, and unforeseen inter-dependencies can wreak havoc on performance. Unravel Data reads existing reporting systems and logs and can insert sensors into the data stack to continuously probe for problems and opportunities to improve performance.

Plan and execute

The software can be used at the planning stage to recommend resource allocation based upon the characteristics of an individual application, Agarwal said. Once in production, it monitors resource consumption and identifies anomalies, such as inefficient queries or inadequate memoriy allocation. For example, “We can detect a degradation of your name node, increased [remote procedure] calls from Impala or Hadoop or if an expensive join is going on,” he said. “We are constantly looking for inefficiencies and errors.”

Improvements in version 4.0 include the ability for the software to detect and diagnose applications that over-allocate resources or under-utilize containers and recommend optimal re-allocation. Unravel Data can now automatically detect and diagnose reasons for application slowdowns based upon SLA definitions and recommend ways to improve performance and reliability. It will alert administrators to bad configuration settings in the cluster and recommend new ones. Improved storage utilization and caching analysis alerts administrators when available storage is reaching capacity and recommends tables and files that can be removed or cached to improve utilization and performance.

The company intends to soon launch support for NoSQL platforms like Apache Cassandra and Apache HBase, Agarwal said. Pricing is based upon the size of the cluster, but the CEO declined to discuss details.

Image: Unravel