Re: [RFC][PATCH 1/6] Storing ipcs into radix trees

[Ingo Oeser]

Hi,

On Friday 08 June 2007, Nadia Derbey wrote:
> Ingo Oeser wrote:
> > ... together with this means 4*256 -> 1k of precious stack space used.
> > Please consider either lowering IPCS_MAX_SCAN_ENTRIES or kmalloc() that.
> You're completely right, but trying to lower the extraction size, I'm 
> afraid this will have an impact on performances.
> 
> Here are the results of a small test I did: I have run ctxbench on both 
> the 256 and and 16 entries versions
> 
> 1) 256 entries:
> 42523679 itterations in 300.005423 seconds = 141743/sec
> 2) 16 entries:
> 41774255 itterations in 300.005334 seconds = 139245/sec

So that is around 1.8% in a benchmark. 

Not bad, if one considers, that this is an expensive syncronisation primitive 
anyway (and thus shouldn't dominate any real workload). At least _much_
better than possible stack underflow :-)

BTW: You forgot to include measurements with the unmodified code 
	as it is in Linus' tree now. They woule be a nice data point here.

> Will try with a dynamic allocation.

But than you have an additional error path 
or have to sleep until memory becomes available.

Maybe try doubling IPCS_MAX_SCAN_ENTRIES - until the performance impact
is in the noise - is simpler. Up to 64 seems acceptable.

Best Regards

Ingo Oeser
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/