Unlimited Event Channels
summary
Adds a FIFO-based event channel ABI that replaces the 4,096-bit table with linked event words, raising practical channels to over 100,000 and providing up to 16 priorities. Implementation uses atomic CAS on link fields and a per-vCPU head copy; tests show uniform average latency across ports at the cost of slightly higher per-event overhead. Xen support is merged and Linux patches target 4.4; a domctl hypercall and a max_event_channels=1023 config option limit guest consumption.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:34How do they work
- 2:24Current layout
- 3:33Problems with current layout
- 4:37Additional requirements
- 7:22Linking events
- 8:41Handling events
- 9:57Event fairness
- 12:39Linux support
- 13:07Priority support
- 16:19Design document
- 16:59Average latency
related talks
34mCoupling Eventdev Usage with Traffic Metering & Policing (QoS)
43mRealtime performance at scale (...)
29mImprove Xeon IRQ throughput with posted interrupt
01h 29mWhen Nanoseconds Matter: Ultrafast Trading Systems in C++
49mKeynote: New Heights
31mBetter Bandwidth Management with eBPF
19mSpeeding Up VM’s I/O Sharing Host's io_uring Queues With Guests
23mHigh Performance Apps with State Channels