As flash arrays become more widely adopted within the enterprise, IT teams are looking for storage products that incorporate predictive analytics to help forecast trends, plan infrastructures and reduce overhead. Cloud-based predictive analytics uses machine learning and other advanced techniques to analyze telemetry data gathered throughout the storage stack, making it possible to proactively address storage issues before they arise.
Predictive analytics is a type of analytics that uses statistical methodologies to forecast specific outcomes based on patterns uncovered in historical and current data. The methodologies can include such advanced techniques as data mining, analytical queries, predictive modeling and machine learning.
Machine learning is particularly important because it automatically trains the predictive algorithms by constantly updating outputs when new data becomes available. As workloads evolve, predictive analytics relies on machine learning to effectively detect and diagnose developing issues and then act upon those issues with minimal intervention.
Predictive analytics and machine learning make it possible to address storage-related issues before they occur, while providing the intelligence necessary to forecast how and when to implement hardware and software changes.
To carry out these operations, a cloud-based predictive analytics product must continuously collect telemetry data from the flash arrays and their supporting infrastructures. The data can include details about IOPS, bandwidth, faults, latencies and other relevant information. The product applies predictive analytics and machine learning to both historical and current data, as it steadily evolves.
Applying predictive analytics to enterprise storage can offer a number of benefits when it comes to forecasting trends, planning infrastructure and reducing overhead. Although these categories overlap, they provide a useful way of understanding the goals of an effective predictive analytics tool.
Because a predictive analytics tool is able to forecast trends, it can take a proactive approach to addressing storage issues, identifying them before they occur and users are affected. The predictive analytics product can then take corrective actions or, if an issue cannot be resolved automatically, notify stakeholders with actionable insights. This approach can help reduce the amount of downtime and increase application availability and performance, while maximizing productivity and the user experience.
The ability to forecast trends requires large quantities of data from multiple systems to properly analyze storage. For example, the Pure1 cloud management service uses the accumulated data from more than 10,000 connected arrays, collecting over 1 trillion data points per day. The service utilizes the Pure1 Meta AI engine to return analytical calculations about the health and functioning of storage and other components in the application stack.
Hewlett Packard Enterprise (HPE) InfoSight provides a similar cloud management service. Built on the Nimble technologies acquired by HPE, InfoSight collects and analyzes hundreds of billions of sensor data points from more than 9,000 customer environments, capturing millions of sensor measurements every second. InfoSight continuously learns from the collected data in order to forecast, prevent and resolve problems. As InfoSight learns, its infrastructure gets smarter and improves, helping to better automate tuning and answer questions before they're even asked.
InfoSight was a key driver behind HPE's decision to acquire Nimble. Real-time predictive analytics also fueled DataDirect Networks' decision to acquire the assets of Tintri this year.
Tintri Analytics, another cloud management service, also relies on a massive pool of telemetry data to deliver a predictive analytics product. The service can analyze up to three years of data from hundreds of thousands of environments in less than a second, incorporating metrics from a number of categories. For example, the telemetry data includes a category for I/O performance, which provides such information as I/O size and read/write throughput. Like Pure1 and InfoSight, Tintri Analytics considers the entire application stack to ensure that forecasts pinpoint the right issues and provide accurate insights.
The predictive analytics capabilities inherent in such services as Pure1, InfoSight and Tintri Analytics also help to more accurately plan infrastructure changes. The services can use the abundance of telemetry data to forecast future capacity, bandwidth and performance requirements and then recommend changes that help prevent future issues, while avoiding unnecessary or risky upgrades.
Tintri Analytics, for example, gathers data at a granular enough level from both compute and storage resources to uncover an application's historical usage patterns. The patterns can then be used to model capacity and performance requirements. Even if an organization is concerned primarily with storage planning, being able to correlate data across the entire application stack can provide a much more accurate picture of future storage needs.
Another example is InfoSight, which builds off its cloud-based predictive analytics capabilities to provide recommendations on how to prevent future issues, optimize resources and improve performance, taking into account I/O workload patterns and other variables that can impact performance. InfoSight continuously refines its machine learning models to provide more precise recommendations, incorporating multiple what-if scenarios into the modeling process.
Pure1 also uses its cloud-based predictive analytics capabilities to forecast capacity and performance requirements over time. Using data from over 100,000 workloads, Pure1 Meta generates analytical profiles of key performance characteristics, while continuously refining the profiles to provide customers with critical, up-to-date insights. For example, Pure1 can provide information into how workloads on a flash array will interact with each other, how capacity and performance requirements will grow over time and whether the array can accommodate additional workloads.
A cloud-based predictive analytics tool can help lower administrative overhead by reducing the amount of time IT teams have to spend troubleshooting issues and coming up with their resolutions. Customers also spend less time calling the support desk because issues are addressed proactively, which also reduces the need for support personnel. All these factors can lead to greater savings, on top of the savings realized by managing systems more efficiently.
According to HPE, for example, customers that utilize InfoSight for their Nimble Storage arrays are benefiting in a number of ways:
- 79% lower storage operational expenses;
- 85% less time spent resolving storage-related troubleshooting;
- 100% of issues going directly to level 3 support engineers, bypassing both levels 1 and 2; and
- 69% faster resolution times spent on level 3 support tickets.
Although these statistics are being reported by HPE, they're consistent with industry-held beliefs that an effective performance analytics tool can help reduce overhead in numerous ways and consequently reduce operational expenses.
Predictive analytics and smarter storage
Applying predictive analytics to storage is still a relatively new phenomenon, but the technology is quickly gaining ground. The techniques used to analyze data and forecast trends are only likely to improve as predictive analytics continues to be adopted.
The ultimate goal, however, is not merely to deliver more accurate and precise analytics, but to implement self-healing infrastructures that autonomously identify and resolve issues with minimal intervention. Predictive analytics, in conjunction with machine learning and other advanced analytical techniques, promises to revolutionize enterprise storage almost as much as the flash arrays.