Revisiting Swapping in User-space with Lightweight Threading

Memory-intensive applications, such as in-memory databases, caching systems, and key-value stores, are increasingly demanding larger main memory to fit their working sets. Conventional swapping can enlarge the capacity by paging out inactive pages backend stores. However, existing solutions suffer several performance compatibility issues, making them unsuitable for high-concurrency memoryintensive applications. In this paper, we redesign system propose Lightswap, a high-performance user-space solution that supports with both local SSDs remote memories. First, avoid kernel involvement, leverage extended Berkeley Packet Filter (eBPF) handling page faults in user space further eliminate heavy I/O stack help of drivers. Then, co-design fault lightweight thread (LWT) scheduling improve throughput reduce end-to-end latency. Finally, try-catch framework Lightswap deal swap-in errors which have been exacerbated scaling process technology. We implement our production-level evaluate it various benchmarks. Results show achieves scalable notification latency (4μs under 128 LWTs), reduces 3-5 times, improves memcached more than 40% compared state-of-art systems.