[Date Prev][Date Next][Thread Prev][Thread Next]
- Subject: Re: GC param tuning
- From: Mike Pall <mikelu-0411@...>
- Date: Fri, 12 Nov 2004 19:00:01 +0100
Adam D. Moss wrote:
> Okay, /8) *9 instead. It's the principle. :) I was aware of
> the overflow, but the worst that happens (I think) is that GC
> becomes over-aggressive rather than bursting into flames.
I think there are two ways the GC gets into trouble, depending on the
direction of the overflow: doing too much in one step (top bits of
threshold cut off) and not doing enough in one step (lim being negative
after the first calculation in luaC_step).
The former may lead to long latencies which is a show-stopper in some
applications. The latter may lead to unwarranted out of memory conditions.
Neither sounds too attractive.