[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: following up on speed

On Wed, 17 Jul 2002, Noel Welsh wrote:

> - for some uses (most often real time) GC is considered bad (ignoring
> research on deterministic GC).  Apparently a reason C++ doesn't
> include GC is so it can be used in military systems that forbid GC
> (again no reference).

I'm currently trying to see what it would take to get my company to start
using the boehm collector for one of our C++ programs (this is what we
consider a spare-time research project).  One of the biggest objections I
expect to have to deal with is that our application typically deals with
large datasets, and we need all the address space we can get.  If, as
reported in the Zorn paper, the boehm collector requires 1.5x - 2x the
address space in order to manage the same amount of actual data, then even
if that address space doesn't get mapped to physical memory it will be a
serious obstacle for us.

Obviously the solution would be to split the data into more than one
process and/or go 64-bit.  We do that when we can, but there are
considerations which prevent either of these from being a possibility in
some cases.

Kimberley Burchett
Endeca, Inc