Here are some numbers from the venerable "kernbench" (just kernel compilation)
to indicate how dcache_rcu (currently in -mm) helps us in the larger systems -
A 16-CPU NUMA-Q running make -j 24
2.5.40-mm2 [includes dcache_rcu]
--------------------------------
Elapsed: 19.622s User: 193.366s System: 42.498s CPU: 1201.6% [average of 5]
2.5.40-mm2-no_dcache_rcu
------------------------
Elapsed: 20.47s User: 193.226s System: 47.548s CPU: 1175.8% [average of 5]
IOW, a 10.6% reduction in system time.
In general, we see significant gains with multiuser workloads in
larger systems.
I also did some "kernbench" testing on UP, 2-CPU and 4-CPU configurations
on a PIII xeon based system. The results are included below and they
show that there isn't any regression at the lower end of systems.
Thanks
-- Dipankar Sarma <dipankar@in.ibm.com> http://lse.sourceforge.net Linux Technology Center, IBM Software Lab, Bangalore, India.---------------------------------------------------------------------------- UP :
2.5.42 vanilla --------------
real 6m12.917s user 5m48.164s sys 0m22.677s
real 6m11.102s user 5m48.103s sys 0m22.635s
real 6m11.170s user 5m48.003s sys 0m22.753s
real 6m11.556s user 5m48.018s sys 0m22.728s
real 6m11.137s user 5m48.185s sys 0m22.592s
2.5.42 dc12 -----------
real 6m12.965s user 5m48.261s sys 0m22.596s
real 6m10.964s user 5m48.059s sys 0m22.539s
real 6m11.513s user 5m48.295s sys 0m22.402s
real 6m11.139s user 5m48.040s sys 0m22.685s
real 6m11.050s user 5m47.969s sys 0m22.720s ----------------------------------------------------------------------
2-CPU :
2.5.42 vanilla --------------
real 3m28.108s user 6m17.968s sys 0m29.230s
real 3m25.086s user 6m17.686s sys 0m29.307s
real 3m24.832s user 6m17.966s sys 0m28.946s
real 3m24.733s user 6m17.737s sys 0m29.302s
2.5.42 dc12 -----------
real 3m26.774s user 6m18.109s sys 0m29.218s
real 3m25.381s user 6m17.892s sys 0m29.155s
real 3m25.719s user 6m17.708s sys 0m29.260s
real 3m25.777s user 6m18.054s sys 0m28.967s
----------------------------------------------------------------------------
4-CPU :
2.5.42-vanilla --------------
real 1m47.632s user 6m25.166s sys 0m34.480s
real 1m48.780s user 6m25.250s sys 0m34.384s
real 1m47.451s user 6m24.915s sys 0m34.644s
real 1m48.709s user 6m25.256s sys 0m34.688s
real 1m47.797s user 6m25.193s sys 0m34.288s
real 1m47.586s user 6m24.901s sys 0m34.587s
real 1m47.452s user 6m25.178s sys 0m34.272s
real 1m47.969s user 6m25.285s sys 0m34.129s
2.5.42-dc12 -----------
real 1m47.638s user 6m25.450s sys 0m33.799s
real 1m47.602s user 6m24.897s sys 0m34.379s
real 1m48.559s user 6m25.188s sys 0m33.941s
real 1m47.945s user 6m25.345s sys 0m33.986s
real 1m47.639s user 6m25.519s sys 0m33.855s
real 1m48.976s user 6m25.051s sys 0m34.189s
real 1m47.447s user 6m25.383s sys 0m34.225s
real 1m47.448s user 6m25.102s sys 0m33.980s
- 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/
This archive was generated by hypermail 2b29 : Tue Oct 15 2002 - 22:00:51 EST