Messages in this thread | | | Date | Fri, 16 Mar 2007 14:23:56 -0700 | From | Paul Jackson <> | Subject | Re: Summary of resource management discussion |
| |
Herbert wrote: > looks good to me, except for the potential issue with > the double indirection introducing too much overhear
It's not the indirection count that I worry about.
It's the scalability of the locking. We must avoid as much as possible adding any global locks on key code paths. This means: 1) be reluctant to add them to fork/exit 2) just RCU locks on per-job (or finer grain) data when on the normal page allocation path 3) nothing outside the current task context for the normal task scheduling code path.
A global lock on the wrong code path is fatal for scaling big NUMA boxes.
... now whether or not that is an issue here, I don't claim to know. I'm just worried that it could be.
Atomic data, such as global counters, is just as bad.
-- I won't rest till it's the best ... Programmer, Linux Scalability Paul Jackson <pj@sgi.com> 1.925.600.0401 - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |