Re: [Gems-users] Different numbers of instructions, msgs, etc when running Ruby on the same workload

[Mike Marty <mikem@xxxxxxxxxxx>]

Lide Duan wrote:
> Oh... so if some msgs finish their traverses more quickly due to the
> different routing algorithm, they might affect the following behaviors of
> the application, thus generating different amount of msgs in different runs,
> right?
>
> If so, how can I make a fair comparison between the performances using
> different routing algorithms? CPI?
>
>   
If you are running multihreaded workloads, I recommend seeing the 
following papers for a discussion of using a work-based metric rather 
than CPI/IPC:
http://www.cs.wisc.edu/multifacet/papers/ieeemicro06_ipc.pdf
http://www.cs.wisc.edu/multifacet/papers/hpca03_variability.pdf

--Mike