Tech
What are the storage requirements for AI training and inference? | Computer Weekly
Despite ongoing speculation around an investment bubble that may be set to burst, artificial intelligence (AI) technology is here to stay. And while an over-inflated market may exist at the level of the suppliers, AI is well-developed and has a firm foothold among organisations of all sizes.
But AI workloads place specific demands on IT infrastructure and on storage in particular. Data volumes can start big and then balloon, in particular during training phases as data is vectorised and checkpoints are created. Meanwhile, data must be curated, gathered and managed throughout its lifecycle.
In this article, we look at the key characteristics of AI workloads, the particular demands of training and inference on storage I/O, throughput and capacity, whether to choose object or file storage, and the storage requirements of agentic AI.
What are the key characteristics of AI workloads?
AI workloads can be broadly categorised into two key stages – training and inference.
During training, processing focuses on what is effectively pattern recognition. Large volumes of data are examined by an algorithm – likely part of a deep learning framework like TensorFlow or PyTorch – that aims to recognise features within the data.
This could be visual elements in an image or particular words or patterns of words within documents. These features, which might fall under the broad categories of “a cat” or “litigation”, for example, are given values and stored in a vector database.
The assigned values provide for further detail. So, for example “a tortoiseshell cat”, would comprise discrete values for “cat” and “tortoiseshell”, that make up the whole concept and allow comparison and calculation between images.
Once the AI system is trained on its data, it can then be used for inference – literally, to infer a result from production data that can be put to use for the organisation.
So, for example, we may have an animal tracking camera and we want it to alert us when a tortoiseshell cat crosses our garden. To do that it would infer the presence or not of a cat and whether it is tortoiseshell by reference to the dataset built during the training described above.
But, while AI processing falls into these two broad categories, it is not necessarily so clear cut in real life. It will always be the case that training will be done on an initial dataset. But after that it is likely that while inference is an ongoing process, training also becomes perpetual as new data is ingested and new inference results from it.
So, to labour the example, our cat-garden-camera system may record new cats of unknown types and begin to categorise their features and add them to the model.
What are the key impacts on data storage of AI processing?
At the heart of AI hardware are specialised chips called graphics processing units (GPUs). These do the grunt processing work of training and are incredibly powerful, costly and often difficult to procure. For these reasons their utilisation rates are a major operational IT consideration – storage must be able to handle their I/O demands so they are optimally used.
Therefore, data storage that feeds GPUs during training must be fast, so it’s almost certainly going to be built with flash storage arrays.
Another key consideration is capacity. That’s because AI datasets can start big and get much bigger. As datasets undergo training, the conversion of raw information into vector data can see data volumes expand by up to 10 times.
Also, during training, checkpointing is carried out at regular intervals, often after every “epoch” or pass through the training data, or after changes are made to parameters.
Checkpoints are similar to snapshots, and allow training to be rolled back to a point in time if something goes wrong so that existing processing does not go to waste. Checkpointing can add significant data volume to storage requirements.
So, sufficient storage capacity must be available, and will often need to scale rapidly.
What are the key impacts of AI processing on I/O and capacity in data storage?
The I/O demands of AI processing on storage are huge. It is often the case that model data in use will just not fit into a single GPU memory and so is parallelised across many of them.
Also, AI workloads and I/O differ significantly between training and inference. As we’ve seen, the massive parallel processing involved in training requires low latency and high throughput.
While low latency is a universal requirement during training, throughput demands may differ depending on the deep learning framework used. PyTorch, for example, stores model data as a large number of small files while TensorFlow uses a smaller number of large model files.
The model used can also impact capacity requirements. TensorFlow checkpointing tends towards larger file sizes, plus dependent data states and metadata, while PyTorch checkpointing can be more lightweight. TensorFlow deployments tend to have a larger storage footprint generally.
If the model is parallelised across numerous GPUs this has an effect on checkpoint writes and restores that mean storage I/O must be up to the job.
Does AI processing prefer file or object storage?
While AI infrastructure isn’t necessarily tied to one or other storage access method, object storage has a lot going for it.
Most enterprise data is unstructured data and exists at scale, and it is often what AI has to work with. Object storage is supremely well suited to unstructured data because of its ability to scale. It also comes with rich metadata capabilities that can help data discovery and classification before AI processing begins in earnest.
File storage stores data in a tree-like hierarchy of files and folders. That can become unwieldy to access at scale. Object storage, by contrast, stores data in a “flat” structure, by unique identifier, with rich metadata. It can mimic file and folder-like structures by addition of metadata labels, which many will be familiar with in cloud-based systems such as Google Drive, Microsoft OneDrive and so on.
Object storage can, however, be slow to access and lacks file-locking capability, though this is likely to be of less concern for AI workloads.
What impact will agentic AI have on storage infrastructure?
Agentic AI uses autonomous AI agents that can carry out specific tasks without human oversight. They are tasked with autonomous decision-making within specific, predetermined boundaries.
Examples would include the use of agents in IT security to scan for threats and take action without human involvement, to spot and initiate actions in a supply chain, or in a call centre to analyse customer sentiment, review order history and respond to customer needs.
Agentic AI is largely an inference phase phenomenon so compute infrastructure will not need to be up to training-type workloads. Having said that, agentic AI agents will potentially access multiple data sources across on-premises systems and the cloud. That will cover the range of potential types of storage in terms of performance.
But, to work at its best, agentic AI will need high-performance, enterprise-class storage that can handle a wide variety of data types with low latency and with the ability to scale rapidly. That’s not to say datasets in less performant storage cannot form part of the agentic infrastructure. But if you want your agents to work at their best you’ll need to provide the best storage you can.
For airlines to run critical operations on networks that are set up and run for them, removing the complexity and cost of managing connectivity themselves, air industry tech firm SITA has launched a new network solution designed to support the demands of complex airport and transport environments.
With around 2,500 customers, SITA technology supports more than 1,000 airports and more than 19,600 aircraft worldwide. The company said that it also helps more than 70 governments “strike the balance between secure borders and seamless journeys” and connects 45-50% of the industry’s data exchange to enable complex global networks to operate smoothly and reliably.
As part of the latter aim, the SITA Campus Network, powered by HPE Aruba Networking, aims to offer a managed network service covering more than 150 countries wherein SITA takes care of the design, procurement, shipping, installation, configuration and support for all devices involved. Boasting a low total cost of ownership (TCO), SITA is proposing “one of the most competitive” fully managed local area network/wireless local area network (LAN/WLAN) available in the industry.
Explaining the rationale for the launch, SITA noted that managing networks across multiple locations, devices and suppliers is complex and costly. Furthermore, it said that when networks are fragmented, performance suffers and disruptions can spread quickly.
SITA Campus Network is attributed with being able to remove this burden by delivering a fully managed network across wired and wireless environments. The campus network is claimed to combine “robust” connectivity with centralised, cloud-based management to ensure consistent, reliable performance across airport campuses and other large transport hubs.
Designed for high-density environments such as terminals, hangars and airline operations centres, the solution is said to support large volumes of users and devices without compromising performance, even during peak demand. By integrating HPE technology into its managed service, SITA’s customers get a network that is centrally operated by SITA while retaining the flexibility to use different technologies and vendors.
Available in more than 145 countries, with 24/7 operational support, SITA assured that by reducing the need for costly hardware and simplifying operations the network lowers both upfront investment and ongoing costs. Its pay-as-you-go model allows customers to scale usage up or down based on demand, with rapid deployment across locations.
This is said to reduce the need for on-site support, spare equipment and recurring training, freeing up IT teams to focus on higher-value activities. Where needed, the campus network connects to SITA’s global wide-area network services. This connectivity links more than 600 airports worldwide.
As is the norm with other leading networking solutions, the SITA Campus Network uses AI to improve visibility across the network, detect issues earlier and automate troubleshooting, helping reduce downtime. It also provides centralised management, allowing infrastructure and devices to be monitored and controlled across both on-site systems and remote environments.
Martin Smillie SITA senior vice-president of communications and data exchange, said integrating diverse systems and devices across airport environments is becoming more complex as operations become more connected: “At the same time, expectations on performance, resilience and security continue to rise. With SITA Campus Network powered by Aruba, we take on that complexity. We deliver a network that is set up, run and continuously optimised, so our customers can focus on keeping operations moving while maintaining control across increasingly demanding environments.”
Sujai Hajela, executive vice-president and general manager for enterprise campus and branch at HPE, added: “Airports and airlines have to support thousands of staff, passengers and mission critical systems across terminals, gates and airside areas – and any network issue shows up immediately as delays and frustration.
“SITA Campus Network powered by HPE Aruba Networking is built on our secure, AI-native technology to deliver a self-driving network that spots and fixes problems in real time, often before anyone notices, so operations keep moving and passengers stay connected.”
Tech
Chinese hackers using compromised networks to spy on Western companies, says Five Eyes | Computer Weekly
China-linked hackers are using networks of vulnerable internet-connected devices, including home routers, printers and smart devices, as cover to mount espionage and hacking operations.
The technique is now used by the majority of China-linked hackers as a way to obscure hacking and espionage attacks launched against organisations in the West.
The UK’s National Cyber Security Centre (NCSC) and national agencies in nine other countries have warned today that Chinese-linked groups are now leveraging networks of infected devices “at scale” to target critical sectors globally and steal sensitive data.
According to an advisory issued by the Five Eyes intelligence-sharing alliance – comprising the UK, the US, Canada, Australia and New Zealand – and 10 other countries, Chinese groups are exploiting security vulnerabilities in unpatched internet devices to create networks to use as a staging post to launch further attacks.
“We know that China’s intelligence and military agencies now display an eye-watering level of sophistication in their cyber operations,” said NCSC chief Richard Horne in a speech at its CyberUK conference in Glasgow.
Covert networks hide ‘indicators of compromise’
The agencies warn that the Chinese tactics are making it difficult for organisations to detect and attribute malicious attacks on their computer networks using traditional “indicators of compromise”.
Chinese groups, for example, could use a UK-based infected device as a staging post to hack into a UK-based company, meaning that blocking non-UK IP addresses no longer provides a defence for overseas attacks.
They advise companies to adopt “adaptive, intelligence-driven measures” to better mitigate the risks, including monitoring traffic from internet-connected devices, virtual private networks (VPNs) and remote access devices to identify suspicious traffic.
Chinese-linked groups are able to evade detection by exploiting low-cost networks of infected devices that can rapidly be reconfigured so that traditional static IP block lists are no longer effective.
The networks are used for each phase of a cyber attack, from reconnaissance and malware delivery, to command and control and data exfiltration against targets of espionage and offensive cyber operations, according to the advisory.
Covert networks behind major hacking operations
Covert networks of compromised devices have been used by the Chinese state-sponsored group Volt Typhoon to pre-position for future attacks on critical national infrastructure (CNI).
The group has targeted communications, energy, transport and water services in the US, and has been able to maintain covert access to critical IT systems for five years or more.
It used a network of vulnerable Cisco and NetGear routers, which were no longer supported by the manufacturers and were no longer receiving updates of security patches.
Another Chinese group, Flax Typhoon, has used a covert network of 260,000 compromised devices, including routers, firewalls, webcams and CCTV cameras, to conduct cyber espionage against targets in multiple countries.
Hacking as a service
Chinese hacking groups have a choice of covert networks, each with potentially hundreds of thousands of endpoints, which frequently change, making it more difficult for companies targeted to block attacks, according to the advisory.
Chinese information security companies have maintained networks of infected devices, available as a service for Chinese-linked hacking groups.
Chinese company Integrity Technology Group controlled a network known as Raptor Train, which infected more than 200,000 devices worldwide in 2024.
Companies advised to take countermeasures
The NCSC advises companies to map internet-connected devices in their organisation and corporate VPNs, so they can understand which traffic is legitimate.
They should also introduce multifactor authentication (MFA) when employees use remote connections to dial into business networks.
Larger organisations can profile incoming connections based on operating systems, time zones, and the organisation’s systems configurations to identify legitimate traffic.
The Five Eyes and the NCSC advise the most at-risk organisations to actively track Chinese advanced persistent threats (APTs), using threat reports supplied by the NCSC to create dynamic block lists and rules to detect incoming threats.
“In recent years, we have seen a deliberate shift in cyber groups based in China utilising these networks to hide their malicious activity in an attempt to avoid accountability,” said Paul Chichester, NCSC director of operations. “We call on organisations to act now to better defend their critical assets.”
Chirp reinvented the wheel—or at least one type, the yoga wheel. Chirp Wheels are effective in relieving upper and lower back pain, sciatica, and tension headaches. WIRED contributor Hannah Singleton has said the Chirp Wheel XR-3 Pack has even helped undo her tech neck and alleviate her brain fog.
Recently, the wellness brand has expanded beyond its flagship wheels into recovery gear. The lineup now includes powered rolling massagers (which I’ve been using a lot lately for back pain relief), TENS units, and even a full massage table (Chirp Contour) that I’m currently testing (stay tuned for the full review). Where Chirp stands out from heavyweights like Hyperice and Therabody is in its simplicity and value. The products tend to focus on doing one thing well rather than piling on features you may never use. Chirp promos and discounts run frequently on the Chirp website, and we have Chirp discount codes, so you can get an even better deal on recovery gear that’s already reasonably priced.
Save up to 67% on Chirp Products With Daily Deals
I like checking Chirp’s Daily Deals page because the exclusive offers rotate frequently, and you can save as much as 67%. I’ve spotted the Chirp Wheel XR 3-Pack on there, but you’ll also find different versions of the wheel, along with storage accessories. Some wheels skip the pressure-point nodes, which can feel better if you’re focusing on improving spinal mobility and flexibility rather than digging into stubborn knots. If the Chirp RPM Mini pops up at a special discount, it’s worth considering for your first purchase. It’s essentially an electric roller that kneads muscles more gently than most percussive massage guns; it also comes with a carrying case, so you can toss it in a bag and take it with you.
Get a Free Chirp Wheel+ Deep Tissue 2-Pack When You Spend $99 or More
Spend $99 or more, and Chirp will throw in a complimentary Chirp Wheel+ Deep Tissue 2-Pack, which retails for $75. The bundle includes two wheels: a 6-inch Deep Tissue Wheel designed for larger muscle groups and a 4-inch Focus Wheel meant to target trigger points in the neck and other small areas. You’ll need to sign up for the email newsletter to claim the freebie before adding it to your order.
Get Free Shipping on Chirp Orders Over $75
Chirp customers receive free shipping on U.S. orders over $75, and the perk stacks with the brand’s daily deals and most codes. If you time it right, you can shave a decent chunk off the final price. No promo code at checkout required.
Chirp Discount Code: Select Customers Can Get 15% Off
Chirp offers a 15% discount to certain groups through an online verification process. That includes: active-duty military personnel, veterans, and their dependents; first responders and law enforcement officers; medical professionals and healthcare workers; and teachers and academic administrators at any grade level.
Italy’s OVS’ FY25 sales rise 7% to $2.06 bn; beats market
Comcast beats revenue, earnings expectations as broadband losses improve
Meghan Markle plots royal bombshell sequel to Prince Harry’s big hit
-
Fashion1 week agoFrance’s LVMH Q1 revenue falls 6%, shows resilience amid Iran war
-
Entertainment1 week agoIs Claude down? Here’s why users are seeing errors
-
Sports1 week agoPSL 11: Peshawar Zalmi win toss, opt to field first against Quetta Gladiators
-
Tech1 week agoThe Deepfake Nudes Crisis in Schools Is Much Worse Than You Thought
-
Business1 week agoStandard Life buys rival in £2b deal to create savings giant
-
Politics1 week agoIran in continuous message exchange with mediator Pakistan after US talks: Report
-
Tech1 week agoCYBERUK ’26: UK lagging on legal protections for cyber pros | Computer Weekly
-
Fashion1 week agoWhat no one is saying about the 2026 apparel slowdown