RCU+SMR Not!

Joe Seigh

unread,

Sep 7, 2005, 7:45:07 PM9/7/05

to

I just found out that there's a patent application in on SMR hazard
pointers so they're not really free to use. At least as far as
using it for FOSS. So RCU+SMR is not anymore.

--
Joe Seigh

When you get lemons, you make lemonade.
When you get hardware, you make software.

Alexander Terekhov

unread,

Sep 8, 2005, 5:41:17 AM9/8/05

to

Joe Seigh wrote:
>
> I just found out that there's a patent application in on SMR hazard
> pointers so they're not really free to use. At least as far as
> using it for FOSS. So RCU+SMR is not anymore.

Even a granted patent would not make it so. BTW, how is the proof of
your perpetum mobile doing? (I mean "barrier free" hazard pointers.)

regards,
alexander.

Joe Seigh

unread,

Sep 8, 2005, 7:37:02 AM9/8/05

to

Looks ok to me. I haven't received feedback from the people who were
supposedly looking at it. I was planning to wait for 4 way or 8 way
processors to become dirt cheap so I could demonstrate scalability and
robustness. But it's a moot point now.

I think the way to go on lock-free is with Java and C# which has a
documented memory model and proper atomicity guarantees. The whole
issue of how to implement a truly concurrent and efficient GC becomes
"somebody else's problem". See how easy that was to fix? :)

John Hickin

unread,

Sep 8, 2005, 8:27:21 AM9/8/05

to

"Joe Seigh" <jsei...@xemaps.com> wrote in message
news:ndGdnbf2cus...@comcast.com...

> I just found out that there's a patent application in on SMR hazard
> pointers so they're not really free to use. At least as far as
> using it for FOSS. So RCU+SMR is not anymore.
>

I guess the patent application's merit is strongly dependent on when the
application was delivered to the patent office for consideration and when
the topic first appeared in any public forum.

Joe Seigh

unread,

Sep 8, 2005, 9:08:17 AM9/8/05

to

Patent application number is 20040107227 filed Dec 3, 2002. Earliest publication
is about Jan 28, 2002. There may have been a provisional patent filed but at
any rate I'm pretty sure IBM and the patent attorneys they hire are aware of
patent law requirements at that time.

Joe Seigh

unread,

Sep 8, 2005, 9:10:15 AM9/8/05

to

Joe Seigh wrote:

> Alexander Terekhov wrote:
>
>> Even a granted patent would not make it so. BTW, how is the proof of
>> your perpetum mobile doing? (I mean "barrier free" hazard pointers.)
>>

[...]

>
> I think the way to go on lock-free is with Java and C# which has a
> documented memory model and proper atomicity guarantees. The whole
> issue of how to implement a truly concurrent and efficient GC becomes
> "somebody else's problem". See how easy that was to fix? :)
>

Well, maybe not. I forgot. Sun has a patent on using GC for lock-free
in Java.

Markus Elfring

unread,

Sep 10, 2005, 1:21:13 PM9/10/05

to

> Patent application number is 20040107227 filed Dec 3, 2002. Earliest publication
> is about Jan 28, 2002. There may have been a provisional patent filed but at
> any rate I'm pretty sure IBM and the patent attorneys they hire are aware of
> patent law requirements at that time.

Would you like to refer to this document?
Method for efficient implementation of dynamic lock-free data structures with safe memory
reclamation
http://freepatentsonline.com/us-app20040107227.html

Does a special application of "hazard pointers" specify a state of the art technique?
http://en.wikipedia.org/wiki/Prior_art

Regards,
Markus