Data is the life-blood of artificial intelligence and machine learning (AI and ML). As these technologies mature and applications proliferate, they will generate vast amounts of data – and with them, new storage challenges.
Organisations need to balance storage performance, ease of management and cost.
That means designing a storage strategy to support AI and ML applications using the optimal storage technologies for the kinds of data AI and ML create. In nearly all cases, that means object storage as a key component of the storage strategy. Why? Let’s look at the reasons.
An AI Future Needs Unlimited Storage Scale
Large datasets are required to train AI and ML algorithms to deliver accurate decisions. This, in turn, drives significant storage demands. For example, Microsoft required five years of continuous speech data to teach computers to talk, and Tesla is teaching cars to drive with 1.3 billion miles of driving data. Managing these data sets requires storage systems that can scale without limits.
After the AI algorithm is trained, it will start generating its own data. The original data set will expand and improve through use. For that to happen, data must be given context through metadata. But humans can’t manually add context to each piece of data; the sheer amount of data would take weeks or months for a human to analyse. Artificial intelligence systems, however, can process such amounts of data in a matter of minutes. Thus, the use of AI to improve AI will further boost demand for data storage scalability.
It does little good to store data sets if you can’t quickly find the data you need. Searchability, powered by metadata, is what makes large volumes of data useful.
In AI and ML, metadata is key to extracting value from data. Object storage allows the data to be described with an unlimited set of tags to make finding specific items within the set easier. It also allows information about unstructured data to be abstracted, a requirement for its application in analytics.
AI and ML learn from many different data types, which require varying performance capabilities. As a result, systems must include the right mix of storage technologies – a hybrid architecture – to meet the simultaneous needs for scale and performance. A homogeneous approach will ultimately fall short.
For data sets that grow without limits, a parallel-access architecture is essential. Without it, the system will develop bottlenecks that limit data growth. Additionally, vast data sets will sometimes require hyperscale data centres with purpose-built server architectures. Other deployments may benefit from the simplicity of pre-configured appliances.
Creating and gathering AI-scale data sets can take years, meaning that losing them isn’t an option. But backing up enormous sets in one go can be costly and time-consuming. Instead, some object storage solutions come with self-protecting capabilities that mean a separate backup process isn’t necessary. These solutions give customers a choice when it comes to the level of protection, enabling users to strike a balance between cost and data protection.
Data locality and Cloud Integration
While some AI/ML data will reside in the cloud, much of it will remain in on-premises data centres for reasons including performance, cost, and regulatory compliance. But, to be competitive, on-premises storage must offer the same cost and scalability benefits as its cloud-based counterpart.
Regardless of where data resides, integration with the public cloud will be an important requirement for two reasons. First, although a lot of AI/ML innovation does occur on-premises, much is also happening in the cloud. So, cloud-integrated on-premises object storage systems will provide the greatest flexibility to leverage cloud-native tools. Second, we are likely to see a fluid flow of data to and from the cloud as information is generated and analysed. An on-premises solution should have the capability of simplifying the flow between the two environments instead of limiting it.
Storage systems geared towards AI and ML systems must be both scalable and affordable, two attributes that don’t always co-exist in enterprise storage. Historically, highly-scalable systems have been more expensive on a cost/capacity basis. Large AI data sets are not feasible if they break the storage budget. Object storage systems are often built on industry-standard server platforms, resulting in a cost-effective solution.
Storage Choices – the Case for Object Storage
These requirements mean that any workable storage strategy for AI and ML will need to include object storage, because that technology offers advantages for AI and ML applications. Chief among them is its ability to scale limitlessly within a single namespace. What’s more, capacity can be added at any time to cater to a growing data set, all whilst being built on the lowest-cost hardware platform, overcoming the traditional cost penalties imposed by large-scale storage.
Additionally, object storage offers metadata and hybrid architecture capabilities, natively integrates with cloud environments, and provides built-in redundancy, meaning there is no need for a separate backup process.
Organisations that want to remain competitive in a future shaped by AI and ML must understand that data will be their biggest asset in future success. Learning from that data history will feed the AI engine tomorrow, but only if the data can be stored, accessed and properly understood today.
This article is from the CBROnline archive: some formatting and images may not be present.