Reference counting garbage collection
erno-news at erno.iki.fi
Fri Aug 24 01:19:25 CEST 2001
In article <7xbsl7a167.fsf at ruckus.brouhaha.com>, Paul Rubin
<phr-n2001 at nightsong.com> writes:
| These days, a typical workstation is a 1 Ghz Athlon with
| 50 nsec memory access time. So today's CPU is > 100x faster (more
| parallelism in addition to higher frequency) but memory is only
| slightly faster. So the cycle cost of a cache miss has gone up
| tremendously--you can execute 50 or more instructions in the time it
| takes to read something from DRAM on a cache miss. This -has- to have
| some effect on the cost of memory fragmentation.
on the other hand it is also worth keeping in mind that back then
there were no caches in cpus, whereas todays cpu caches as big as main
memories were then (and see very good hit rates).
More information about the Python-list