3 ms·
Benchmarking boot latency on x86
- kabdib 10y agoGenerally speaking, POST will dominate over OS boot time on modern server hardware. POST is on the order of minutes, tens of minutes for machines with terabytes of memory . . . and that's with the "fast boot" option selected. I fondly remember the days of desktop systems with their OSes in ROM. You could do "Reset, user interface" about as fast as you could say the words. The whole POST stack has lost its way.
- drewg123 10y agoThe sooner we can get rid of both legy BIOS and EFI in favor of Coreboot or something like it, the better. My son's x86 based Chromebook boots to a login screen from cold power off in just a few seconds, and I've always attributed it to using Coreboot rather than BIOS. I mostly don't care about boot times (which is saying a lot, as I do FreeBSD kernel development and reboot server class systems dozens of times a day sometimes), that's just a minor benefit of Coreboot. The legacy nature of BIOS and the complexity of EFI both freak me out from a security / stability point of view. And don't even get me started on the crappy server board vendor's java kvm / impi tools. Sigh. The whole crappy ecosystem is due for a disruption.
- amluto 10y agoThis article makes the assumption that the TSC is zero at power on and counts monotonically from then on. The problem is that BIOS vendors know this and sidings try to game it. I have a machine in which it appears that the BIOS resets the TSC partway through POST, probably to improve benchmarks. (I can tell this is happening because they only reset it on one core, which breaks all kinds of things.)
- yuhong 10y agoI remember the advice to disable the HPET when Ryzen was released, which also reminds me of this: https://lkml.org/lkml/2015/6/3/4 https://lkml.org/lkml/2015/6/3/4