From: Dave Abrahams (abrahams_at_[hidden])
Date: 2000-01-30 10:54:11
on 1/30/00 10:46 AM, Gavin Collings at gcollings_at_[hidden] wrote:
> "greg colvin" <gcolvi-_at_[hidden]> wrote:
> original article:http://www.egroups.com/group/boost/?start=1991
>> Could you send us the results in some non-proprietary format -- I don't have
>> Excel and don't want to buy it.
> Yes, sorry about that. I really added it as an afterthought in case
> any one wanted to play around with the numbers. I should have included
> the full results in text form though, so here they are (in ns per op).
> They broadly agree with other tests posted since. Each copy operation
> was, on average, half an assignment and half a copy construction.
> Shared Linked
> Initial Construction 2900 +/- 70 -30 +/- 50
> Copy Operation 62 +/- 13 119 +/- 9
> Then the figure 2900 / (119 - 62) = 50 (or worst case 22 at the
> extremes of the error bands) gives the number of copy operations needed
> before the increased copy time for linked pointer outweighs the initial
> overhead of allocating the reference count on the heap.
I really think it's time to trot out your new implementation which uses
deque so we can test against that (I'm thinking the allocation cost will be
much lower since the deque works like a pool allocator). If you need help
finishing it, please call on me.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk