2026-07-21 –, Room 1
"Something is wrong with the network. I used to get 4Gbps transfers but now I'm only getting 120Mbps. Did you change something recently?"
Sound familiar? If you've spent any time supporting production systems, you've probably heard some variation of this complaint. Before jumping to conclusions about where the problem lies, we need to understand what's actually happening at the TCP layer on both endpoints.
This presentation is inspired by a series of articles I wrote on LinkedIn focused on practical, real world troubleshooting of application network performance issues. Instead of deep diving in to packet captures, this presentation will highlight the various tools and metrics available on Linux systems that help engineers and administrators pinpoint the true source(s) of application performance problems when "the network" is blamed.
We will focus on how packets are processed in a Linux system from the NIC, through the OS, up to the application. We will then take a closer look at TCP itself, utilizing native Linux tooling to introspect on individual TCP socket performance. This TCP socket introspection will dive in to each socket's calculated network latency (RTT), packet sizing (MTU and MSS), and packet loss (retransmissions, SACK, duplicate ACKs
Principal Network Engineer | Low-Latency Infrastructure | Global Datacenter Architecture
With extensive experience in network architecture and infrastructure, I currently serve as a Principal Consultant at Kyberis Networks. My work focuses on leveraging expertise in Networks, Systems, and Security to address complex challenges and deliver innovative technical solutions tailored to organizational needs.
Previously, I contributed to high-speed, low-latency network design and operations at a financial trading company, where I applied advanced knowledge of IP network architectures and datacenter technologies. My passion lies in enabling secure, scalable, and efficient systems that support mission-critical business objectives.