Proposed agenda items:
1. Tailored instances of Xen: continuing the Nov 2018 discussion of KCONFIG/L0 hypervisor use cases. More details upcoming via wiki page.
2. Macro supply chains: what are best practices for maintaining Xen macros which originate in other open-source communities, e.g. QEMU or Linux? Would each macro benefit from a documented status (e.g. "Ignore upstream changes", "Monitor upstream changes", "Mirror upstream changes") with associated tooling?
3. Go toolchain: is there community interest in collaborating on the development of golang tools for local management of Xen? Historically, OpenXT used a combination of Haskell and Ocaml tools. Some OpenXT community members are using golang with Xen. Could these new tools find a home in upstream Xen?
4. NVME passthrough performance: this is improved when VMEXITs are avoided by using "posted interrupts" [1] available on Broadwell and later Xeon processors or AWS nested hypervisor "metal" [2]. For commodity x86 CPUs which do not have posted interrupts, Linux [3] and Hyper-V [4] have used "hybrid polling" to achieve good I/O performance at the cost of CPU cycles. Is this applicable to Xen?
Rich
[1] "VT-d Posted Interrupts" - Intel, 2012
[2] Running Thousands of KVM Guests on Amazon i3.metal Instances, twosix, 2017
[3] "I/O Latency Optimization with Polling" - Western Digital, 2017
[4] "Achieving 10-Million IOPS from a single VM on Windows Hyper-V" - MS, 2018
