NICA: An Infrastructure for Inline Acceleration of Network Applications
summary
NICA builds an ikernel abstraction and runtime that runs AFUs on a Mellanox FPGA+ASIC board, prototypes on Mellanox Anova Flex, and integrates with VMA and KVM. It implements fine‑grained time sharing, per‑VM packet tagging, and weighted packet schedulers for performance isolation. A memcached ikernel adds 107 lines of code and reaches 40 Gbps (99% line‑rate) at 6 μs 99th‑percentile for 16‑byte items — 21× a 6‑core CPU — with linear scaling to multiple VMs.
AI-written summary · about
chapters (13)
- 0:00Start
- 1:07Inline Acceleration (Analogy)
- 1:33Inline Processing - Filtering
- 2:25COAP Cryptographic Authentication
- 4:01The ikernel Abstraction
- 4:22Attaching an ikernel
- 7:56Cloud Deployment Model
- 10:50Evaluation
- 12:02Memcached Cache
- 13:51Key-Value Store Cache - Virtualization
- 14:16Latency under virtualization
- 14:58Key-Value Store - Performance Isolation Throughput over time
- 15:55Conclusion
related talks
15mThe nanoPU: A Nanosecond Network Stack for Datacenters
20mNetKernel: Making Network Stack Part of the Virtualized Infrastructure
14mHigh-throughput and Flexible Host Networking for Accelerated Computing
17mSmartNIC-Enabled Live Migration for Storage-Optimized VMs with PYROCUMULUS
22mTransparent MultiNIC Routing for Large AI Models
21mAlNiCo: SmartNIC-accelerated Contention-aware Request Scheduling for Transaction...
16mLemonNFV: Consolidating Heterogeneous Network Functions at Line Speed
19mFpgaNIC: An FPGA-based Versatile 100Gb SmartNIC for GPUs