Subject: Re: [boost] Proposal: Monotonic Containers - Comparison with boost::pool, boost::fast_pool and TBB
From: Simonson, Lucanus J (lucanus.j.simonson_at_[hidden])
Date: 2009-06-22 18:16:52
Christian Schladetsch wrote:
> I do not suggest, nor have I ever suggested, that monotonic is a
> general-purpose allocator that is a drop-in replacement for
> in all cases. I have repeatedly stated the intended limititation of
> monotonic allocation, including mentioning that it is not designed for
> continual erasure from and insertion into containers. But it does
> what it
> was designed for very well, which is to produce fast and small
> datastructures with very fast allocation.
Yes, I understand your intention. You want to be able to use a scratchpad memory with standard containers. I think the idea has merit. I want to understand what happens when the allocator is misapplied. Clearly if people use it as it was intended to be used they should get the benefits you intended them to get. I want to know what happens to everyone else. It looks like my theory about cache and working set size is not borne out by your testing.
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk