Who doesn’t like the idea of having six 9’s of uptime for their storage? Availability is a major consideration for any storage, public or private. Of course, having great availability isn’t the only measure of storage quality. We also value performance, low latency and consistency. But making storage behave in a predictable manner isn’t always easy. In recent years, storage vendors have been keen to tell us about their ability to build insight from data collected in the field. This insight translates into actions that minimise downtime and otherwise keep our storage arrays ticking along nicely.
Nimble Storage (now an HPE brand/company) has been collecting metrics since the first arrays shipped on 15 April 2010. That’s seven years of customer data from hardware in the field. Nimble claims 4000 individual data points are retrieved every 5 minutes from each system, including the hardware and file system. The granularity is per second, with some 30-70 million pieces of data collected from an array each day. Automated resolution is an impressive 86%, with 54% of issues being attributed to problems outside of the array itself. External issues include configuration and integration problems with storage networking, hosts, and applications.
Some of the information InfoSight produces is relatively straightforward, such as capacity planning data. However, the real benefit is when the collective set of data from all customers is analysed to highlight problems other customers might have not yet experienced. This collective knowledge is a step up from the previous best practice where storage arrays all had the ability to call home. Array vendors implemented some predictive diagnosis, but the rate of data collection was never as detailed and as frequent as it is today.
The latest step forward for InfoSight is to use machine learning and artificial intelligence to identify and resolve issues that wouldn’t otherwise be obvious to a human being. HPE is claiming they are the first to achieve this. However, I do remember the AI message already being promoted by Pure Storage at Accelerate earlier this year (see this post). Kaminario also claims their Clarity analytics tool has “application-level intelligence” and leverages “machine learning” (link) – this from a press release in September 2016. So it looks like machine learning is already widespread. Now the devil could be in the detail here. Exactly what the intelligence is, how the actions are taken and what the results are, aren’t entirely clear. Chris Mellor highlights just this point in his article on the new InfoSight news:
There are no actual numbers to show how the new AIRE InfoSight will improve things; bettering the six “nines” for example, by making it seven.
Lest we think it just airy-fairy stuff, HPE has corralled a quote from Justin Giardina, CTO for iland Secure Cloud, who said: “The new recommendation engine is phenomenal as it’s making proactive decisions, showing us how we can improve our environment.”
That’s all very well, but until we see real numbers testifying to its effectiveness we can’t assume AIRE InfoSight is anything other than promising tech.
Whilst I agree with Chris, the lack of information isn’t entirely true. Nimble has presented at many recent Storage Field Days (part of the Tech Field Day series). There are lots of InfoSight video examples online that demonstrate the value of the analytics process. However, as yet, none of them show how machine learning/AI takes this process a step further. This is what Chris was looking to highlight.
Retrofitting Insight for 3PAR
InfoSight now covers the 3PAR platform and will be available early next year for systems running 3PAR OS 3.3.1 GA and later. This is good news, but of course one of the benefits highlighted by Nimble before acquisition was the collection of data from day 1. In fact (see image 1) Nimble marketing always claimed InfoSight was an “architectural advantage that can’t be retrofitted”. Obviously, HPE has since found a way to make that possible with 3PAR. To be fair, if 3PAR systems were already collecting data, then the advantage of InfoSight is the depth of the analytics being performed, rather than collection per se.
The Architect’s View
It’s easy to criticise shared storage as a thing of the past, but if you’re looking to get high availability with low impact, then features like InfoSight are essential. The ability to collect information from many systems and have a much greater sample of data means problems spotted in one customer can be avoided in another. This translates to real operational time saving and reduced business impact. Collective knowledge is something that simply doesn’t exist in software-defined storage.
I hope HPE has historical data from 3PAR systems, otherwise, it may be a while before the benefits of InfoSight are realised. Nimble aren’t likely to do Tech Field Day again, so we’ll have to look to HPE Discover to see exactly how the power of AI translates into improved system resilience.
- Pure1 META – Analytics for Pure Storage Arrays
- Kaminario Unveils Cloud Platform to Deliver Application-Level Intelligence to All-Flash Storage (Kaminario press release, retrieved 21 November 2017)
- Gulp. HPE’s InfoSight self-repairs and makes ‘proactive decisions’ (The Register, retrieved 21 November 2017)
- Tech Field Day Companies – Nimble Storage (Tech Field Day website, retrieved 21 November 2017)
- Powered by HPE InfoSight, the future is autonomous (Around the Storage Block blog, retrieved 21 November 2017)
Comments are always welcome; please read our Comments Policy first. If you have any related links of interest, please feel free to add them as a comment for consideration.
Copyright (c) 2007-2020 – Post #DDB7 – Brookend Ltd, first published on https://www.architecting.it/blog, do not reproduce without permission.