Trend #1 Artificial Intelligence (AI) – AI is one of the most over-hyped terms in the industry. In 2019 that hype reaches new heights. Vendors will not only claim they have the perfect storage system for AI workloads but that they are leveraging AI to improve the customer experience.
The reality is that most organizations are nowhere near close to implementing anything along the lines of true AI. At best organizations are using machine learning, and most are using analytics. Machine learning workloads and analytic workloads can undoubtedly benefit from higher performance storage systems.
Organizations need to quantify if the investment in extra storage performance makes a difference for them. Factors to consider are; applications need to be able to take advantage of these speeds, there has to be enough data to process, and there has to be a benefit, typically a competitive advantage, to the organization in processing that data more quickly. Without question, some organizations have workloads that meet these qualifications, and their number is growing, but most organizations are not there yet.
The organization’s place on the AI/ML journey dictates what type of storage system they should purchase. Organizations that are well beyond simple analytics processing and are moving toward machine learning should look at high-performance storage systems and file systems explicitly designed to meet that need. Traditional storage systems should serve organizations that are still doing batch reporting and basic analytics well. The key is not to end up with a dozen different storage systems, each with a designated workload. In the modern data center, some amount of storage sprawl is inevitable, but IT needs to look for solutions that can serve multiple workloads to minimize sprawl wherever possible.
Another component of the AI hype is vendors claiming to use AI to make their storage systems or networks smarter and easier to use. While, again, most of these use cases are for faster analytics processing than they are AI or even machine learning, there is a practical need for smarter, more automated storage and network infrastructures.
There are examples of this automation already today. A cache analyzes data patterns and decides which data should be in the cache and which data should be on regular storage. The problem is that once data moves out of cache, because of inactivity, applications or users need to access that data several times before the dataset is promoted back into the cache again. The result is while the data is reestablishing its importance users suffer a drop in performance. A machine learning algorithm might realize that a particular data set becomes very active at certain times of the year. Once it realizes that, it may pre-promote that data to the cache, eliminating the potential for a cache miss.
More advanced automation leverages analytics and machine learning across all the vendor’s customers to spot trends or potential problems. Nimble Storage, now owned by HPE, started this trend with its InfoSight and HPE has quickly broadened the use case across much of its product offering. Other companies are following Nimble’s lead.
The reality is the most AI is more fiction than reality. The bulk of work done in the space is at best machine learning but more likely fast analytics processing. AI is still a work in progress that has a future in almost every organization. Until that future is a reality, organizations need systems the deliver fast analytics processing for their applications. Even shorter term is leveraging systems that provide advanced analytics to help make storage systems efficient.
Our next blog looks at Hyperconverged Infrastructure (HCI). 2019 should be a big year for HCI, and it should consume more workloads than in years past, potentially moving it out of the point solution category and into broad data center adoption.