On 24.02.24 07:41, Mark Blackman wrote:
If an Amazon firecracker environment works for you, there’s
https://www.freebsd.org/status/report-2023-04-2023-06/boot-performance/
Oh. Thanks. 20 ms boot time sounds good enough to me :-D
.. aww, the PDF link to the slides is broken :-(
https://wiki.freebsd.org/BootTime (Colin Percival) works and reports ~8
seconds on FreeBSD 14.0.
http://www.daemonology.net/blog/2021-08-12-EC2-boot-time-benchmarking.html
(again, Colin Percival 2021) tells me that the best boot-to-TCP time was
about 1.23 s, while typical boot times would be around 10-15 s.
Those 8 seconds are actually good, unless the competition has sped up as
well :-)
(This benchmarks just TCP availability, I don't know or how much needs
to be added for ssh availability.)
There's a conclusion here: While those improvements are awesome, 8
seconds are nowhere near the do-not-disrupt-developer-workflow
threshold, so the VM snapshot it will be.
Oh. Wait. These links talk about FreeBSD's boot time. I don't know how
this relates to Firecracker.
OTOH Firecracker is very new, so I'm somewhat reluctant to jump that
bandwagon anyway.
> of course, there’s more to booting than the kernel.
Definitely. In my own testing, something that vaguely sounded like a
mail subsystem had a full minute-long wait.
Some tweaking will be needed to get rid of that behaviour. I'm
comfortable with that, though not with having to put those tweaks into a
setup script and keeping it up-to-date with every new FreeBSD version. I
haven't seen an OS or distro that does not have this kind of problem
though, so I'll just have to live with that.
I’d guess some of those improvements could apply to more generic VM hypervisors
too.
Probably, but as much as I like exploring rabbit holes, I already have a
too-long list of these to add yet another one :-)
Regards,
Jo
