Re: Wait Free Proxy Collector
- From: Chris Thomasson <cristom@xxxxxxxxxxx>
- Date: Tue, 29 Jan 2008 14:40:45 -0600
On Tue, 29 Jan 2008 06:01:42 -0800, Dmitriy Vyukov wrote:
On Jan 29, 3:19 am, "Chris Thomasson" <cris...@xxxxxxxxxxx> wrote:
There was one proxy thing I did but never published which used an N=2
scheme
Yeah, but Paul E. McKenney was published it. And he didn't forget to
fill in patent too :)
I'm talking about his real-time RCU with 2 counters per processor.
Well, it seems like that flavor of RCU still cannot reference count an
arbitrary number of references outside of RCU protected regions. Try and
to virtually zero-overhead distributed reference counted message passing
with RCU as-is...
;^)
, sounds like your doing same thing here. It did not use atomics, but
it did require a single #StoreLoad barrier before a thread could enter
a proxy collected region. I was trying to come up with atomic-free, but
not membar free, proxy designs to test against vZOOM PDR to see what
kind of impact the membar has. Turns out, the overhead can be quite
significant indeed.
I bet you:
1. store some per-thread data
2. allow arbitrary amount of completely unused garbage hanging around
and "waiting for next epoch"
If yes, then it's already more like RCU then PC.
Yeah; you right. Humm.
.
- References:
- Wait Free Proxy Collector
- From: ti_chris
- Re: Wait Free Proxy Collector
- From: Dmitriy Vyukov
- Re: Wait Free Proxy Collector
- From: ti_chris
- Re: Wait Free Proxy Collector
- From: Chris Thomasson
- Re: Wait Free Proxy Collector
- From: Dmitriy Vyukov
- Wait Free Proxy Collector
- Prev by Date: Re: atomic-free allocator
- Next by Date: Re: atomic-free allocator
- Previous by thread: Re: Wait Free Proxy Collector
- Next by thread: Re: Wait Free Proxy Collector
- Index(es):
Relevant Pages
|