Re: lockstat from pathological builds Re: high sys time, very very slow builds on new 24-core system

[Andrew Doran <ad%netbsd.org@localhost>]

On Sat, Apr 02, 2011 at 01:03:13AM -0400, Kevin Bowling wrote:

> Is anyone working on lockless implementations within NetBSD?
> 24-core systems are becoming very common, even in two socket
> pizzabox rack servers.
> 
> RCU is one of the foundations of the Linux kernel but is patent
> encumbered with a GPL exception by IBM.  There are may be valid
> alternatives using something like epoch observation.  Check out this
> BSD licensed project
> http://concurrencykit.org/doc/safeMemoryReclamation.html for a
> rundown and good implementations.
> 
> I'm not sure how realistic widespread implementation would be as a
> GSoC project but I'd be willing to investigate some limited scope if
> there is a suitable mentor available.

We have code for interesting & low overhead synchronization methods like RCU
but it hasn't gone in yet, I'll let others say more if they want to.  That
said, there's already a ton of lockless stuff going on in the kernel, just
no song and dance made about it.

If you look at the lockstat output from Thor's system there are only a
couple of problem areas, the most significant being management of a central
set of data structures in the VM system.  There is no silver bullet like RCU
for that, it needs more involved work.  On the positive side, the rest of
the system scales up just fine on this workload.