- people - netdevconf.info

0x1A: Talk, Kernel shared memory socket transport
by Jamal Hadi Salim 24 Jun '26

24 Jun '26

High performance IPC between two local apps uses zero copy on shared memory. There's a plethora of APIs that require a plethora of libraries' support - and both the sender and receiver MUST use the same IPC library to communicate. sigh. In this talk, Davide Wei proposes adding kernel support for zero-copy shared memory IPC (both sender and receiver) built on AF_UNIX SOCK_SEQPACKET + io_uring, enabling any userspace library to interoperate without sharing a common IPC library. An RFC prototype is ready for community feedback on uAPI and design. More details: https://netdevconf.info/0x1A/sessions/talk/kernel-shared-memory-socket-tran… cheers, jamal

1 0

0x1A: Talk, Line-Rate Cybersecurity: Modern DPI and Encrypted Traffic Fingerprinting at 100 Gbps
by Jamal Hadi Salim 23 Jun '26

23 Jun '26

Luca Deri and Alfredo Cardigliano present recent advancements in nDPI, an open-source Deep Packet Inspection toolkit, addressing challenges posed by encryption and evasion protocols that limit legacy firewalls. The talk covers using cryptographic fingerprints to identify malicious actors despite encryption, and exposes structural flaws in JA3/JA4 fingerprinting methods when handling ephemeral TLS extensions. Finally, Luca and Alfredo will detail integrating nDPI with the Linux kernel firewall for real-time traffic optimization, plus architectural blueprints using PF_RING and SmartNIC flow managers to achieve deterministic 100 Gbps monitoring and hardware-accelerated enforcement. Come and interact with Luca and Deri! More details: https://netdevconf.info/0x1A/sessions/talk/line-rate-cybersecurity-modern-d… cheers, jamal

1 0

0x1A: Talk, Thrice the charm: an skb extension for BPF metadata
by Jamal Hadi Salim 22 Jun '26

22 Jun '26

Jakub Sitnicki revisits storing custom BPF data tied to an skb's lifetime. In this talk Jakub will recap what he's already established on the XDP/skb metadata feature (scope, accessible hooks, wipe conditions, and driver contract) and then discuss two approaches for paths where XDP metadata isn't supported: a BPF map side-stash available today, and an RFC for embedding a BPF metadata buffer in an skb extension chunk. Jakub will compare both from the BPF-hook interface and overhead perspectives, and explain why alternatives (extending metadata lifetime, BPF local storage) failed. The latter will delve into skb extension internals—compiling out unneeded extensions and avoiding hot-path allocations—and finally the talk closes with open user-API questions: whether a new socket option or cmsg type is needed, how users exchange metadata today and its limits, and how metadata maps to L7 abstractions like streams and requests. Come and engage Jakub! cheers, jamal

1 0

0x1A: Two talks on offloading security to DPUs
by Jamal Hadi Salim 22 Jun '26

22 Jun '26

In this talk Balakrishna Bhamidipati and Vijay Ram Inavolu describe total offload of TLS from the host onto a DPU. They run a reverse-proxy on the NIC to handle MCP traffic. This moves TLS termination, OAuth2/JWT validation, and session-aware L7 routing off the host and onto the NIC. The authors will describe the packet path, kTLS activation and fallback, session lifecycle, backend affinity, and SSE relay, showing how the design can be reproduced on commodity DPUs without any kernel modifications. https://netdevconf.info/0x1A/sessions/talk/macsec-protected-rdma-on-dpus-fr… In the second talk Alkama Hasan and Vijay Ram Inavolu describe how they resolve the hard problem of security for zero-copy mechanisms like RDMA. In such approaches, the payload bypasses the network stack altogether leaving the app outside the security fence that typically protects ordinary host networking. Large-scale AI and HPC jobs depend on RDMA to move data between nodes. MACsec is a natural fit to resolve this security challenge at layer 2. So what does such a DPU approach offer above the NICs that support MACsec offload today? Well, merely offloading MACsec lacks the ability to interact with a multi-tenant cluster orchestrator which a DPU-based approach can. IOW, an orchestrator (Kubernetes in this case) has no visibility into which nodes actually offer MACsec-protected RDMA egress, so a workload can't request placement on such a node. The authors illustrate this making MACsec become a schedulable property exposed in the node ResourceSlice via DRANet-Sec allowing workloads land on the right nodes automatically. https://netdevconf.info/0x1A/sessions/talk/macsec-protected-rdma-on-dpus-fr… cheers, jamal

1 0

0x1A: Talk, Can Homa and TCP Get Along?
by Jamal Hadi Salim 20 Jun '26

20 Jun '26

John Ousterhout examines TCP and Homa coexistence. Initial measurements show that when running TCP and Homa concurrently, TCP performs slightly better than Homa because Homa reduces its buffer utilization. However, over time Homa degrades badly because TCP selfishly fails to reciprocate and overloads buffers. John introduces homa_qdisc, a queuing discipline that paces both protocols' traffic, implements SRPT for Homa and limited SRPT for TCP, and balances output during congestion. Results? with homa_qdisc, both protocols improve. When running concurrently—Homa suffers only slight degradation, and TCP latency improves. Even better: TCP on its own rips the benefits of the homa_qdisc: with tail latency for short messages nearly halving versus fq_codel. Come and engage with John! cheers, jamal

1 0

0x1A: Talk, AF_XDP copy mode needs more love
by Jamal Hadi Salim 19 Jun '26

19 Jun '26

In the spirit of the legend Bob Marley (playing while i am sipping coffee and typing this!): AF_XDP: Could you be loved? Jason Xing is showing a lot of love to you. Most drivers can't do zero-copy, so copy mode is the universal path. Jason will talk about recent correctness fixes (multi-buffer TX leaks, continuation descriptors, TOCTOU metadata) cleared the bugs, while fine-grained profiling unlocked close to 2× throughput. Questions for you AF_XDP (yes, you!): Could you be locked - with finer grain? Could you be batched - without the pain? No single silver bullet! Just cycle-level perf, contention tracing, cache-line heatmaps. And the optimization love for you: - Send without budget blues - Hot structure reshuffles (cache-line dance) - Batch xmit — one ring to rule them - Doorbell coalescing (don't ring it twice) - Skb allocation acceleration - IRQ pairs trimmed short Don't let them fool ya — spinlock's not the ruler. Methodical, mon, methodical, one cache line at a time, thats how we get there! Ok, enough of my spin, read more details on the talk here: https://netdevconf.info/0x1A/sessions/talk/af_xdp-copy-mode-needs-more-love… cheers, jamal

1 0

0x1A: Talk, Networking Headless CXL Devices for AI Memory Services
by Jamal Hadi Salim 19 Jun '26

19 Jun '26

Another talk on the intersection of AI and memory requirements and how networking is a fundamental piece of the equation. AI workloads needs large, high-bandwidth, low-latency memory. Composable CXL memory fabrics address this by moving compute to data, running services such as search, caching, and quantization on smart Type-2 devices with local DDR. The trade-off? Device-side side needs IP endpoints, while CXL devices expose a memory window, not a NIC. In this talk, Vijay Inavolu and Gaurav Agarwal present a Linux virtual-interface path built from stock kernel pieces. Host and device daemons open /dev/net/tun, configure a virtual L3 interface, and mmap a shared CXL HDM-H window as a packet ring, enabling local apps to access memory with no kernel changes. The shared-ring design, CXL ordering rules, and a new host-pod bridge pattern give device-side Linux a cluster-facing service identity, so services are discovered and reached by service IP while clients stay unaware of the memory window. Measurements show 60x less host-link traffic than host-side compute and 2.56x FAISS throughput scaling across four cards. Vijay and Gaurav would like to find partners in this crime to build composable AI memory fabrics on top of Linux networking. Come, listen, provide feedback and yes collobarate! cheers, jamal

1 0

0x1A: Talk, The CXL Fabric End-Game: Bandwidth Realities and Networked Memory for AI Scale
by Jamal Hadi Salim 19 Jun '26

19 Jun '26

AI/ML workloads have a memory problem. The model memory requirements are growing exponentially but the GPUs just cant keep up. No worries; CXL makes gobbles of memory available. The RAM can be disaggregated and placed in racks outside the system reachable via a network link. PJ says CXL.mem is typically pitched as a memory expansion technology (treating it like a "far" NUMA node), but this framing understates a critical issue: CXL link bandwidth can be orders of magnitude lower than native DDR, not just higher latency. As AI workloads push toward distributed memory pools, this bandwidth gap fundamentally changes the cost model for scaling large models across disaggregated memory. PJ has some suggestions for upcoming CXL 2.0 to move beyond simple expansion toward true memory-coherent cluster execution, calling for coordinated work in switching infrastructure, firmware, and Linux kernel networking/memory subsystems. Come, listen and provide feedback to PJ! cheers, jamal

1 0

0x1A: Talk, Linux QUIC: Bringing a Modern Secure Transport into the Kernel
by Jamal Hadi Salim 18 Jun '26

18 Jun '26

Xin Long proposes migrating QUIC from user space into the kernel. He introduces QUIC as a native transport via a new IPPROTO_QUIC socket type enabling direct use by subsystems such as SMB and NFS. The architecture handles transport logic in the kernel while offloading TLS handshakes to user space, supporting POSIX-style APIs. The implementation has been validated through large interoperability tests and benchmarking. Come learn about this shiny new subsystem! More details: https://netdevconf.info/0x1A/sessions/talk/linux-quic-bringing-a-modern-sec… cheers, jamal PS:- Today is the last day for early bird registration i.e 50% off!

1 0

0x1A: Talk, TCP State of the union
by Jamal Hadi Salim 18 Jun '26

18 Jun '26

Eric Dumazet, aka "network ninja" will give the classical 2026 TCP State Of The Union. Recent and upcoming TCP changes to address performance scaling on modern platforms will be part of the highlights. Dont miss it! cheers, jamal

1 0

0x1A: BoF, Next generation Intel Ethernet Drivers
by Jamal Hadi Salim 17 Jun '26

17 Jun '26

Sridhat and JP will give us the innards of the nextgen Intel Ethernet drivers. ixd and idpf are Intel Ethernet drivers built around a common networking IP, spanning multiple product generations and form factors — discrete PCIe NICs, SoC NICs, SoC switches, and IPUs. idpf is an open-specification, unified PF/VF driver implementing the IDPF PCIe programming interface. It has been upstream in the Linux kernel for several years, is shipped in x86/ARM distributions, and is widely deployed in Google Cloud. ixd is a newer driver that combines control and data path functionality and leverages the switchdev framework to provision, configure, and manage devices that expose idpf-based data path functions. The BoF will cover the architecture and design of both drivers, the challenges of supporting diverse device modes, and the strategy for refactoring idpf to share code between ixd and idpf. It will also explain the different PCI function types (PF/VF) and subfunctions exposed by the hardware, and how the drivers are loaded to support the various possible configurations. Come and participate! More details: https://netdevconf.info/0x1A/sessions/bof/next-generation-intel-ethernet-dr… cheers, jamal

1 0

0x1A: BoF, Network Observability
by Jamal Hadi Salim 17 Jun '26

17 Jun '26

Jason Xing will lead a BoF on the exciting topic of network observability. Network observability has evolved from classic tools like `tcpdump` toward richer, layered capabilities that let engineers trace root causes after problems arise. The scope of this BoF is to discuss new ideas and survey existing techniques (which collect wide-ranging data points such as latency, throughput, skb drops, protocol diagnostics, reference counting, and many others). Specific sub-topics on the current agenda: 1. BPF timestamping 2.0: finer-grained, very low-overhead timestamping for broader use cases. 2. IETF TCP reset diagnostic payload: replacing current reset reasons with common APIs and messaging. Very welcom topics for discussion include how AI's rapid evolution may reshape network observability. Please ping Jason if you want to participate! More details: https://netdevconf.info/0x1A/sessions/bof/network-observability-bof.html cheers, jamal

1 0

0x1A: Talk, Resilient AI Supercomputer Networking using MRC and SRv6
by Jamal Hadi Salim 16 Jun '26

16 Jun '26

At very large scale AI training, tail latency and not average latency, is what limits synchronous pretraining performance. Christoph Paasch will discuss how MRC resolves these issues by: 1. Introducing a new RDMA transport protocol that sprays traffic across many paths and actively load-balances, eliminating flow collisions. 2. Creating Multi-plane Clos topologies and enabling >100K-GPU training clusters as two-tier networks with high switch radix and physical redundancy. 3. And last, using static SRv6 source-routing which gives MRC autonomous, failure-bypassing path control. The talk will describe experiences running MRC and static SRv6 routing in production within OpenAI and Microsoft’s largest training clusters where it has been used to train the latest frontier models. Christoph will discuss operational experience showing how MRC allows AI training jobs to ride out many network failures that previously would have interrupted training. cheers, jamal

1 0

0x1A: Bronze Sponsor, Common Net!
by Jamal Hadi Salim 16 Jun '26

16 Jun '26

We are pleased to welcome Common Net as a bronze sponsor! Common Net is not only a bronze sponsor but is also playing a critical role in hosting the conference in Rome. We'd like to thank for all their efforts in helping make the conference a success. Common Net is an independent Italian telecom operator that owns and operates its own data centers, fiber, and fixed-wireless networks. Founded by engineers, it provides cloud and connectivity services across Italy using open-source and standard technologies, with no resellers or vendor lock-in. It also contributes upstream to open-source projects including the Linux kernel, and is sponsoring Netdev 0x1A in its home city, Rome. More details: https://netdevconf.info/0x1A/news/bronze-sponsor-common-net.html cheers, jamal

1 0

0x1A: Workshop, SRV6
by Jamal Hadi Salim 16 Jun '26

16 Jun '26

Stefano Salsano chairs the classical SRV6 workshop. Current topics for discussion include: - Source-routed AI backend networks - SRV6 L2 services beyond VXLAN - Provider-grade deployment and service protection Come and participate in these discussions! More details: https://netdevconf.info/0x1A/sessions/workshop/srv6-workshop.html cheers, jamal

1 0

0x1A: BoF, Your Fu is Better Than Mine, 3.0!
by Jamal Hadi Salim 15 Jun '26

15 Jun '26

PJ is back with the ever-popular, often-requested, BoF v3! PJ says: This is a safe space, a judgement-free zone. Come with your cool hacks, workflows, test harnesses, whatever. This is a place where all of us who have worked in this space for decades or months can converge, and share knowledge of how we work in this space better. If you’re interested, just bring a short demo. Bonus points this year if we can ride the AI train, and bring a Skill or share effective prompts. Or a simple benchmark that you can show why it’s important to you. No judgements, just pure sharing. Often people in the audience get encouraged by who is presenting, and spontaneously decide to present something of theirs. That’s the spirit of the OG BoF. Let’s do this again, again. cheers, jamal

1 0

0x1A: BoF, Could an IPv6-Only Kernel Be a Reality? An Architectural and Performance Evaluation
by Jamal Hadi Salim 15 Jun '26

15 Jun '26

Fernando Fernandez Mancera chairs this BoF which explores the feasibility of a standalone IPv6 Linux kernel, decoupling it from its current mandatory dependency on the IPv4 stack. The BoF presenters will share an experimental implementation, analyze its impact on architecture and performance, and seek community feedback on an upcoming RFC to identify potential blockers for upstream integration. cheers, jamal

1 0

0x1A: BoF, Need for Bidirectional Event Handlers in Auxiliary Bus
by Jamal Hadi Salim 15 Jun '26

15 Jun '26

The auxiliary bus enables clean PCI device partitioning into multiple drivers but lacks a native signaling mechanism for bi-directional event notifications. Dave Ertman says this restricts communication to a passive model where the parent merely exposes data for the child to consume, hindering real-time coordination for critical events like resets, configuration changes, or cross-driver resource management. Dave has a proposal to address these challenges but there are open questions. Come and interact and provide your feedback. More details: https://netdevconf.info/0x1A/sessions/bof/bof-need-for-bidirectional-event-… cheers, jamal

1 0

0x1A: XDP Workshop!
by Jamal Hadi Salim 14 Jun '26

14 Jun '26

Alexander Lobakin will chair the classical XDP workshop. Come and participate in discussions on current topics and toss some new ideas! cheers, jamla

1 0

0x1A: Tutorial, Making Time Uncertainty a First-Class Concept in Linux Timing
by Jamal Hadi Salim 14 Jun '26

14 Jun '26

Maciek Machnikowski continues his tradition of teaching us about Time. In this session, Maciek reframes timestamps as intervals with quantified uncertainty rather than precise points, arguing that explicit uncertainty windows let applications reason about temporal correctness without hidden accuracy assumptions. He discusses sources of uncertainty such as clock sync offsets, network delay jitter, hardware timestamping limitations, and kernel-to-userspace transfer paths. Ignoring these, argues Maciek, causes errors in ordering, causality detection, and compliance checks. He will describe a time-uncertainty daemon sample and discuss Linux kernel API gaps. Come and grow your knowledge with a real human maestro! More details: https://netdevconf.info/0x1A/sessions/tutorial/making-time-uncertainty-a-fi… cheers, jamal

1 0

0x1A: early bird rate extended to June 18
by Jamal Hadi Salim 13 Jun '26

13 Jun '26

The cheep cheep early bird registration has been extended to June 18. Register now to take advantage of the discount. https://netdevconf.info/0x1A/pages/registration.html cheers, jamal

1 0

0x1A: New bronze sponsor, Mpiric
by Jamal Hadi Salim 13 Jun '26

13 Jun '26

We welcome first time sponsor Mpiric. Thanks for the bronze sponsorship Mpiric. Mpiric Software enhances the security, reliability, and capabilities of the Linux kernel networking stack, with a current focus on MPTCP. cheers, jamal

1 0

0x1A: Bronze sponsor, Red Hat!
by Jamal Hadi Salim 13 Jun '26

13 Jun '26

Much thanks to Red Hat for the bronze sponsorship! Red Hat is an open hybrid cloud technology leader, delivering a consistent, comprehensive foundation for transformative IT and artificial intelligence (AI) applications in the enterprise. As a trusted adviser to the Fortune 500, Red Hat offers cloud, developer, Linux, automation, and application platform technologies, as well as award-winning services. cheers, jamal

1 0

0x1A: Workshop, What's next for the PSP Security Protocol
by Jamal Hadi Salim 12 Jun '26

12 Jun '26

Willem de Bruijn will chair a workshop on PSP Security Protocol. PSP is an inline cryptography protocol optimized for large-scale datacenter deployments. It scales to high performance and high connection count. PSP is supported by network devices from Broadcom, Intel, Nvidia and others and by the Linux kernel as of v6.18. Current hardware implements the initial PSP Architecture Spec. In this workshop Willem will lead a discussion on candidate extensions for the next version of PSP, such as confidential compute and multi-host support. Future directions for the Linux kernel PSP stack and support infra such as packetdrill test coverage will be covered. cheers, jamal

1 0

0x1A: Bronze sponsor, secunet!
by Jamal Hadi Salim 12 Jun '26

12 Jun '26

We are pleased to welcome back friends of the conference, secunet, as a bronze sponsor for Netdevconf 0x1A! secunet is a long time supporter of the conference. We are grateful for their continued support. So who is secunet? secunet(https://www.secunet.com/) is Germany’s leading cybersecurity company. In an increasingly connected world, the company’s combination of products and consulting assures resilient digital infrastructures and the utmost protection for data, applications and digital identities. secunet specialises in areas with unique security requirements – such as cloud, IIoT, eGovernment and eHealth. With security solutions from secunet, companies can maintain the highest security standards in digitisation projects and advance their digital transformation. secunet is an IT security partner to the Federal Republic of Germany and a partner of the German Alliance for Cyber Security. cheers, jamal

1 0

2026

2025

2024

2023

2022

2021

2020

2019

2018

people