Keep Your Data Center Humming With the Power of Big Data – One Data Point at a Time
By Karthik Krishnaswamy – Nimble Storage Product Management
Big Data technologies are being used to analyze large volumes of data across a variety of industries and organizations including healthcare, retail, manufacturing, and even government agencies. We at Nimble Storage are applying big data and predictive analytics principles in a proactive fashion to ensure efficient operation of enterprise data center environments.
How do we do it? Just in the course of a one week period, InfoSight, the Nimble Predictive Analytics platform, analyzes more than a trillion data points! Predictive analytics are then used to correlate vast amounts of information to find the needle in the haystack and proactively suggest resolutions and solve complex issues. By constantly collecting and analyzing data across our entire installed base of more than 8,100 customers, InfoSight is able to easily detect anomalies in the environment, and more importantly, detect patterns and conditions that result in abnormal behavior such as high latency or low throughput. InfoSight proactively safeguards the entire installed base from experiencing issues that could adversely impact business.
Let me provide you with three concrete examples:
Inoculating Against Performance Degradation for 600 Customers: A large educational institution experienced a drop in throughput – from 200 Mb/s to 20 Mb/s — a 10x drop. The reason? The hypervisor was not responding properly to an iSCSI backpressure command resulting in a large number of writes. The Nimble Data Science team was able to immediately identify other customers that could potentially be impacted with this issue by analyzing the workload profiles, and Nimble OS and hypervisor versions. As a result, the customers that fit the similar profile were ‘blacklisted’ from upgrading to the specific version of Nimble OS that would expose the hypervisor issue, therefore avoiding performance degradation. These blacklists allow our customers to “sidestep” interoperability issues by avoiding problematic hypervisor and storage OS versions. InfoSight also provides users with an optimal path for upgrading Nimble OS on their arrays. This automated blacklisting step allowed 2PB of data to be delivered with absolute performance that may otherwise have taken 10x longer!
Protecting 100 Customers from Malicious Cyber Attack: Some customers configure their arrays with a public IP address leaving them exposed to cyber attacks. InfoSight employs ethical hacking methods to detect whether any data is exposed. If an array is reachable with a public IP address, InfoSight immediately notifies the customer with a P1 case. The array would then be deployed behind a firewall. In this way, InfoSight was able to protect 400 TB of data at-risk of brute force attacks at 100 customer sites.
Root Causing Non-Storage Related Performance Issues: A customer started experiencing sporadic periods of high latency that rendered the system unusable. By performing end-to-end correlation across virtual machines, physical hosts and storage layers, InfoSight identified a specific pattern of network retransmits due to a faulty NIC (Network Interface Card). The customer replaced the NIC, immediately fixing the issue. Neither the server or hypervisor vendor were able to resolve the issue.
In conclusion, by employing predictive analytics, nine out of 10 issues are automatically detected with prescriptive guidance on how to resolve them. As a result, we are able to ensure our customers availability of 99.9997% across thousands of arrays.
In my next blog, I will discuss key capabilities of InfoSight that allows users to monitor and troubleshoot their storage and VM environments.
- Karthik Krishnaswamy