Persistence VM?

Ian Piumarta ian.piumarta at inria.fr
Mon Aug 19 20:26:10 UTC 2002


On Mon, 19 Aug 2002, Stephen Pair wrote:

> Sounds correct to me, but you meant "with one extra word for each
> compact header" instead of "with one extra word for each non-compact
> header" correct?

I meant "for each header that would become non-compact after giving up on
compact headers".

> Also, even if the cci not zero, isn't it possible for that object to be
> larger than 255 bytes (thus requiring a three word header even though
> it's a compact class)?

Absolutely.

> If so, then even 7% is a tad (but just a tad) high.

FWIW:

  ok: 415298 objects in image
  type 0 =   5552 ( 1.34%) size+class header
  type 1 =  84016 (20.23%) class header
  type 3 = 325730 (78.43%) short header
  4736 (1.43%) of compact objects have size header

(type 2 doesn't show up because it's the header on the free chunk) so the
real saving from compact headers is

	.7843 * 415298 * 4 / 1024 = 1272 kilobytes

rather than the 1290 I quoted earlier.

Ian






More information about the Squeak-dev mailing list