Latest report by R&S highlights security and data challenges facing GenAI deployments in networks
ipoque, a Rohde & Schwarz company has announced its latest research report ‘Advancing network management with generative AI: The role of DPI-driven traffic intelligence’. The report finds security risks related to GenAI — such as poisoning of training data and query-based data exposure — being the biggest challenge in the adoption of GenAI, affecting 92.0% of vendors. This is followed by poor mechanisms in data collection, classification and analysis which lead to gaps in the data and analytics used in GenAI. This affects 89.3% of vendors, according to the report.
Network and traffic insights underpin the adoption of GenAI in network management, where machine-generated configurations, simulations, analyses and other outputs help administrators improve network policies and responses. “Accurate network and traffic analytics are critical in training, testing and fine-tuning generative models such as LLMs, GANs and VAEs, as well as in providing references for models such as RAG. These insights are also important in enhancing user prompts and queries during inferencing,” says Ariana Leena Lavanya, Principal Analyst at The Fast Mode.
The report, based on a survey of 75 leading networking vendors, reveals several concerns related to the reliability of existing network and traffic data sources. Inaccurate categorisation and labelling emerge as the biggest concern, followed by disparate and conflicting data. More than half of vendors admit they are impacted by loss of visibility due to new encryption protocols such as TLS 1.3, QUIC, ESNI, and DoX. Other issues cited by vendors include privacy concerns, lack of depth and granularity, incompatible data formats, data lags, and tampered data.
GenAI workloads are resource-intensive, latency-sensitive and attack-prone, increasing the susceptibility of GenAI-based network functions to performance and reliability issues. According to the survey, close to nine out of ten vendors are unable to fully ascertain how network conditions impact their GenAI workloads, and how these workloads, in turn, affect network performance. Similarly, only one in ten vendors are adequately equipped to handle GenAI-related security attacks. “Without full visibility into traffic flows, we anticipate a higher prevalence of unauthorised access, data exfiltration, code errors, and AI-specific threats such as data poisoning, model inversion, and adversarial attacks,” added Ariana.
“Recognising the need for accurate network and traffic insights, ipoque provides networking vendors with next-gen DPI engines, R&S®PACE2 and R&S®vPACE, designed to meet the demand for real-time application and threat awareness,” said Martin Mieth, Director Network Analysis, ipoque. “We integrate advanced classification techniques, encrypted traffic intelligence—combining machine learning and deep learning techniques to identify traffic flows regardless of encryption, anonymisation, or obfuscation—and a rich set of KPIs.”
“Our DPI engines fulfil all three key criteria for GenAI implementations — scalability, real-time analytics, and embeddable into traditional, cloud, and virtualised environments — as identified in the report,” added Martin. The survey reveals that nearly half of vendors have already deployed DPI, with two-thirds expected to do so within the next five years.
Highly customised DPI insights ensure that the data that is used to test, train and fine-tune a GenAI model is aligned with the underlying functions (e.g. traffic compression, QoS monitoring, and traffic filtering) and network types (e.g. (W)LAN, (SD)WAN, and cloud networks). These insights also enable vendors to monitor and optimise their CPU, TPU, and GPU clusters, storage devices, DL frameworks, data lakes, and APIs, ensuring efficient and high-performant GenAI architectures and systems. “ipoque’s DPI comes with support for IPFIX reporting formats, allowing network owners to form a unified source of shared intelligence that helps harmonise GenAI-driven network policies and decisions in multi-vendor environments,” said Martin. Additionally, threat awareness from DPI enables network owners to detect malicious, suspicious and irregular traffic flows, mitigating GenAI-related threats.
The report, conducted in collaboration with The Fast Mode, a leading telecoms/IT publication, delves into various ways advanced network and traffic intelligence powers GenAI-based network functions. It assesses tool options for data gathering and analysis, while highlighting key implementation challenges, including the prevalence of GenAI- threats. Apart from evaluating next-gen DPI’s role in enhancing GenAI deployments, the report also examines its adoption rates and preferred deployment models across vendors.
