How To Avoid Global RAGged AI Delivery Woes....

If Q1 of this year for me was the season of “newbies” – speaking to, and working with, new vendor clients, then Q2 has definitely been the season of “catch-ups”.

The essence of “catch-ups” is the fundamental implication that both parties have been around for some time… In the case of the latest catch-up, however, with Aryaka, here we are talking about a vendor that only dates back to 2009 and my association with them goes back to c.2012. However, such is the pace of change in networking and security since then, Aryaka has effectively already lived through – and grown as a result of – three distinct phases; WAN Optimisation, breaking new ground by delivering it as a service, SD-WAN (still not entirely sure what the difference between those two is) and nowadays SASE or, more specifically, unified SASE, where networking and security take joint top billing.

However, it’s not simply a case of reshaping the company to reflect those IT infrastructure changes, but also to support the latest application crazes. In the case of Aryaka, having its own secure, optimised delivery network has mean adapting to SD-WAN and SASE has come more naturally than with many other vendors in said space(s). But in terms of adapting to adopt the application usage trends, that requires a different kind of focus, and a hint of imagination, in terms of being proactive, rather than reactive. Hence, having created a low-latency platform to cope with real-time application usage surges (think pandemic + homeworking = video craze) Aryaka is now turning its attention to supporting the latest application trend – namely AI. The company has therefore just announced its “GenAI Network Acceleration Solution”, designed to provide the same levels of access to AI workloads (and these can be very, very enormous) as it has previously with video, big data and other bandwidth killer applications.

One trigger for this is the seemingly rapid adoption of RAG (Retrieval Augmented Generation) in the next phase of AI, connecting LLMs (Large Language Models) to legacy applications and data stored across the existing network, which has serious performance and security implications. For those at the heart of AI-related development, such as in the HPC (High Performance Computing) sphere, the understanding of the massive compute resource required to generate series AI-related activity is not necessarily reflected in their understanding of what is then required to transport and access that data globally. For example, since when has a DevOps guy ever worried about how much compute resource they are using (they don’t see the AWS bills)? Thankfully, those of us in the networking world DO understand what is required. Aryaka plans to release its AI>Perform solution over the next two quarters, and can be summarised as providing:

• Optimised Performance: This is via Aryaka’s global core network, the Aryaka Zero Trust WAN, effectively creating a private, low-latency backbone. This backbone is designed to provide predictable performance for AI workloads, regardless of the user’s location, ensuring that AI algorithms can process data rapidly and deliver real-time insights and deliverables.
• Global Reach: For enterprises operating across multiple regions, it is essential to have a solution that can deliver AI workloads efficiently worldwide. This is where the Aryaka network, with its global Points of Presence (PoPs) and network route optimisation comes into play, to maintain that consistent performance and reliability.
• Scalability and Flexibility: As AI workloads grow in complexity and scale, enterprises need a solution that can adapt to their evolving needs. Aryaka deploys a cloud-based single-pass architecture (so every networking and security function is performed only once) which streamlines the expansion of AI activities without creating bottleneck situations and infrastructure constraints.
• Simplified Management: Managing the global delivery of AI applications comes under exactly the same centralised management as existing applications on the Aryaka network; (MyAryaka) provides visibility and control over the entire network and security infrastructure, so the AI workload delivery is completely visible at all times.

Overall, it seems that Aryaka is right on the money again here. From my perspective, as ever, validating that solution would be my “peace of mind” moment 😊 – let’s say “watch this space” on that one…