[nylug-talk] x86-64 for 1GiB+ programs (especially in VMs) -- WAS: LOWMEM vs. HIGHMEM performance advantage?

[Eric Moore]

jh <jhlists at hirschman.net> writes:

>>>> b) some userland application stupidity (O(N^2) algorithm or somesuch)
>>> This is possible, but sadly, out of my hands.
>> do benchmarks on different dataset sizes and see if doubling the dataset 
>> results in quadruple time to process?
>
> Hard to say. For example, one test set consists of 400k "units", and 
> runs in 7 hours. Second test set consists of 400k units, and runs in 
> about 8. Both use around 400ish megs of memory.
>
> Combining both test sets into an 800k unit test set gives us about a 22 
> hour run time, using 900ish megs of memory.
>
> But... the application does shell out to do certain graphic-related 
> operations. It is hard to say how much time is spent in the python app, 
> and how much is spent in something it shells out to.

Looks O(N^2) to me.  Try a 200k unit test set?  I'd guess 3-ish hours.

something like: 0.1875 N^2 + 1.25 * N (N in 100k units)

-- 
Eric