Re: MIPS

On Thu, 12 Jul 2001 16:42:24 BST, "Thakare, Kiran" 
<KThakare(_at_)radioscape(_dot_)com>  said:

> I have a minor question.  How would I calculate the MIPS in order to
> estimate the baseband processing?
> 1.  How to calculate - Formula
> 2. What I need to know know in order to calculate this?
It's totally unclear what it is you're trying to estimate with "baseband
processing".  If you could explain what you're trying to do in more detail,
we might be able to help.

If you're trying to estimate the overhead of your network stack, the best
metric is probably "code path for normal-case packet".  Some kernels
implement zero-copy (which can be a win), and I remember some ancient
history where Van Jacobsen got the normal-case path down to some absurdly
low number like 35 opcodes (any of you other dinosaurs out there have a
reference for this)?

Note that MIPS in general are almost totally useless these days, especially
with multiple execution units on a CPU - even in the 6-year-old PowerPC
architecture, it was possible to effectively complete 4 or 6 instructions
in one clock cycle (if they involved a multiply-add, an unrelated floating
point, an unrelated integer operation, and a branch).  It was also possible
due to pipeline stalls (due to cache miss, data dependency, etc) for the
entire processor to stall for 15 or 20 clock cycles.  The Linux kernel
does compute a 'BogoMIPS' value - but that's ONLY used for "this is how
many times in a given specific null loop to delay 1/Nth of a second".
-- 
                                Valdis Kletnieks
                                Operating Systems Analyst
                                Virginia Tech

pgpOvbMJr3Ihr.pgp
Description: PGP signature