Re: [Myrinet] Performance degradation using gm_register_memory instead of gm_dma_calloc!

On Wed, Jan 23, 2002 at 12:07:55PM +0800, geoge_zhang wrote:
> [...]
>   As is shown in the above result, for small mesages, the performance is largely 
> the same. But for larger messages, the performance of two DMA buffer request approaches
> is quite different.
>   We think the performance degradation of gm_register_memory may result from its more 
> overhead of virtual-physical address transform.
>   Does anyone can explain something more about this topic? Any information will be greatly
> appreciated!

Could you check the alignement of the message buffers in both cases, and try
to use pointers aligned to the same boundary to see if the performance
difference is still there in this case?

Regards,

Loic