[svlug] Inits

Jesse Monroy jesse650 at gmail.com
Sat Jan 17 18:56:29 PST 2015


On 1/17/15, Alison Chaiken <alison at she-devel.com> wrote:
>>
>> Steve Litt wrote:
>> > You have me at a disadvantage because I don't think I've ever seen a
>> > system boot indeterminately, or at least not in the last 10 years.
>
>
> Rick writes:
>> Anyway, truth to tell, I'm likewise at a disadvantage in attempting to
>> intelligently discuss this problem, being both late to the party and
> having
>> through either luck or careful avoidance of problem scenarios not seen
>> the
>> referenced problems.
>
> Steve and Rick, if you look in your dmesg log, you'll find that the order
> of devices coming up at boot under the control of the kernel is quite
> asynchronous.    Kernel developers have a limited ability to sequence
> operations.   If one device driver doesn't call the methods of another, or
> take a reference on the other, they may come up in any old order.   Linux
> has been becoming increasing asynchronous, preemptible and unpredictable
> with each passing kernel.   The advent of systemd is a sign that the trends
> in the kernel are being made visible to userspace.
>
> Akk writes:
>
>> Things may have changed -- this is based on problems I used to hit
>> five or so years ago -- but here are two cases that used to cause
>> indeterminacy:
>>
>
> What really causes indeterminacy?   Hardware.   If only that pesky hardware
> with its variable response times would go away, we wouldn't have all these
> problems.
>
> Why is the Linux kernel becoming so asynchronous?   Because its power
> management and speed used to stink compared to Windows and iOS on the same
> hardware.   Don't take my word for it; look at these Intel slides from
> 2010:
>
> https://events.linuxfoundation.org/slides/2010/linuxcon2010_brown.pdf
>
> Len Brown is a Linux developer employed by Intel.   I see no reason to
> think that his numbers are not valid.
>
> That was 2010.   If Linux's power management were still that bad, Android
> ::::SNIP::::
>

Rick & Steve,

I wanted to say I agree with what Alison has suggested, but from a
different perspective.

Some of you may have noticed that Intel has stopped increasing the
clock speeds on CPUs. John Sokol and I noticed back in 2000.
It was one of those days where I opened my mouth and John said,
PROVE IT. The end result was the benchmarking all machines in the room.
At the time we had i386,  386DX, 486, 486DX, Pentium, Cyrix, AMD,
etc, with the top machine being a "Xeon PII". The results were plain
as day, and my comment proved false, but something close was true.

http://www.dnull.com/cpubenchmark/budmark2.html
http://www.dnull.com/cpubenchmark/buds-O3.html

My statement was, "Intel has done no better, because all they do is
increase clock speed, which has the side effect of less efficiency."

It turns out I missed two factors, hardware BUS speed, and device
latency & response.

Which leads me to predict that the next large increase in computing
will NOT be on the silicon - but in using multiple cores. Which is where
the need for an "event-driven" OS will be.

That's it.
Jesse



More information about the svlug mailing list