Prashant Kumar Singh, Kshitij Gupta and Shivendra Srivastava write about how AWS Lambda replaced an aging network capture system with a new pipeline built using eBPF and Rust. The old architecture relied on kernel-side extensions that faced significant performance issues due to rule explosion in iptables as microVM density increased. By moving packet interception to the kernel via eBPF's traffic control (tc) hooks and handling aggregation in unprivileged, memory-efficient Rust processes, Lambda achieved constant-time lookups, supported IPv6, and significantly reduced CPU/memory overhead without dropping packets or modifying customer data.
- The new system uses a decoupled architecture: eBPF for kernel capture, Rust taggers for userspace aggregation, and an orchestrator to manage the fleet of taggers per host.
- To maintain security, unprivileged Rust processes receive access to ring buffers through file descriptors passed via Unix domain sockets using SCM_RIGHTS.
- The system achieves byte-for-byte identical Amazon Ion records compared to the legacy system, allowing for seamless migration without changing downstream consumers.
- Ring buffer sizes are mathematically derived from peak packet rates and drain intervals (100ms) to ensure no events are lost during bursts.
- The new system uses a decoupled architecture: eBPF for kernel capture, Rust taggers for userspace aggregation, and an orchestrator to manage the fleet of taggers per host.
- To maintain security, unprivileged Rust processes receive access to ring buffers through file descriptors passed via Unix domain sockets using SCM_RIGHTS.
- The system achieves byte-for-byte identical Amazon Ion records compared to the legacy system, allowing for seamless migration without changing downstream consumers.
- Ring buffer sizes are mathematically derived from peak packet rates and drain intervals (100ms) to ensure no events are lost during bursts.
Brendan McKeag writes about Qwen3.8-27B, a dense 27B-parameter vision-language model from Alibaba that achieves frontier-class agentic coding performance while fitting on a single 24GB GPU. The model's hybrid attention stack (48 linear-attention layers against 16 full attention layers) slashes KV cache memory, and its dense architecture makes it far more practical for fine-tuning than the MoE alternatives dominating the industry. McKeag argues the model serves best as an active tool-calling assistant rather than an encyclopedia, and its small footprint makes it uniquely suited for Runpod Serverless scale-to-zero deployments where cold start is dominated by bytes on disk.
- The model is roughly 3x more token-hungry than Qwen3.6-27B, trading wall-clock speed for better answers
- It crossed 1M Hugging Face downloads within days of release; the community nickname is "local Opus"
- Unsloth shipped QLoRA support the day after weights dropped; 4-bit QLoRA is the well-trodden dense route, while MoE QLoRA at 4-bit is explicitly unsupported
- FlashBoot snapshots require loading the model at worker boot before runpod.serverless.start() to be effective
- Harvey and Engram reported it averaging 67% across 250 legal tasks in a synthetic law firm of ~10,000 documents, ahead of every model in that study
- The built-in multi-token prediction head enables speculative decoding without a separate draft model
- The model is roughly 3x more token-hungry than Qwen3.6-27B, trading wall-clock speed for better answers
- It crossed 1M Hugging Face downloads within days of release; the community nickname is "local Opus"
- Unsloth shipped QLoRA support the day after weights dropped; 4-bit QLoRA is the well-trodden dense route, while MoE QLoRA at 4-bit is explicitly unsupported
- FlashBoot snapshots require loading the model at worker boot before runpod.serverless.start() to be effective
- Harvey and Engram reported it averaging 67% across 250 legal tasks in a synthetic law firm of ~10,000 documents, ahead of every model in that study
- The built-in multi-token prediction head enables speculative decoding without a separate draft model
Grafana and GitLab have released a new open-source solution that links GitLab CI/CD events into Grafana's observability stack via a serverless architecture, enabling real-time visibility and correlation between deploy events and performance metrics.
Fly.io provides a secure and fast platform for deploying AI workflows and LLM-generated code using ephemeral, kernel-isolated virtual machines (Fly Machines). It offers features like secure sandboxing, fast startup times, a clean slate for each run, a simple API, and support for whole applications, not just code snippets.
A curated guide to code sandboxing solutions, covering technologies like MicroVMs, application kernels, language runtimes, and containerization. It provides a feature matrix, in-depth platform profiles (e2b, Daytona, microsandbox, WebContainers, Replit, Cloudflare Workers, Fly.io, Kata Containers), and a decision framework for choosing the right sandboxing solution based on security, performance, workload type, and hosting preferences.
OnDemand AI provides API services for media, services, and plugins, allowing developers to upload media, use NLP, and deploy machine learning models. It also facilitates serverless application deployment and allows BYOM (Bring Your Own Model) and BYOI (Bring Your Own Inference).
