Re: Variable-block memory managers bad in realtime

[hbaker@netcom.com (Henry Baker)]

In article <uf90t16x43.fsf@synquiry.com>, Jon S Anthony <jsa@synquiry.com>
wrote:

> gwinn@res.ray.com (Joe Gwinn) writes:
> > Will do.  Do you have more specific references?  Thanks.

> > We are doing the equivalent of engine control in many places.
> 
> If that's really true (with cycle times measured in microseconds),
> this stuff isn't going to work for you.

Well, it depends upon what you mean by 'this stuff'.

Well-engineered dynamic storage algorithms can be used for hard real-time
applications, but they do have to be designed for this purpose.  A number
of 'multi-media' systems with hard real-time requirements have been built
using real-time GC techniques.

Most dynamic storage allocators -- including nearly all implementations
of 'malloc' -- have not been designed for hard real-time use.  Most mallocs
have been designed to make the standard benchmarks run fast (if even that!)
and probably have to be replaced for many serious applications.  There have
been some very interesting measurements of the 'standard' mallocs that ship
with various operating systems, and they aren't pretty.  So the problem
isn't _GC_, per se, but non-real-time allocators themselves.

A number of people in the GC community are interested in embedded systems
applications of GC.  Do a web search for Kelvin Nielsen.