[erlang-questions] Does read-write lock has a better performance than actor model?

[Fred Hebert]

The usual pattern for that kind of thing is to use ETS tables in
protected mode. The server writes to the table, but the interface
functions read directly from it.

There's an example on how to transform the message-based approach to the
ETS-optimized one (when necessary) at
http://learnyousomeerlang.com/ets#a-little-less-conversation

Regards,
Fred.

On 04/15, Xiao Jia wrote:
> I want to propose a possible solution, but I am not confident:
> 
> We probably want to publish readable states to a process (or several
> processes with load balancing?) dedicated to serving read requests.  The
> cost is that after each write operation, the state must be published again.
>  If there are much more reads than writes, the read serving process's
> mailbox size will be much larger than the write serving process.  If I
> remember it correctly, this helps the scheduler to give more priority to
> the read serving process, or to give less priority to readers because the
> mailbox is already full of requests.
> 
> As I said, I am not confident about my proposal.  Hope someone helps to
> correct me :-D
> 
> 
> On Tue, Apr 15, 2014 at 3:46 PM, Bin Wang <wbin00@REDACTED> wrote:
> 
> > I meas if there is a write, then give it a lock. If all the requests are
> > read operations, then no lock is needed. That means multiple read
> > operations could be doing at the same time while there is no write
> > operations.
> >
> > However, in actor model, the requests are always handled one by one no
> > matter they are reads or writes. So didn't it slow at some situations?
> > (Read are more that write).
> >
> > ---
> > Bin Wang
> >
> >
> > 2014-04-15 21:37 GMT+08:00 Dmitry Kolesnikov <dmitry@REDACTED>:
> >
> >> Hello,
> >>
> >> No, this is not good.
> >>
> >> Any on-going "write request" makes changes to the state. The read request
> >> does not know when the state becomes consistent unless there is a lock on
> >> the state. You cannot ensure safe reading while writes is going on.
> >> Therefore, actor model has bigger advantage.
> >>
> >> - Dmitry
> >>
> >>
> >> On 15 Apr 2014, at 16:31, Bin Wang <wbin00@REDACTED> wrote:
> >>
> >> > In Erlang, we handle concurrency in such a way: a process receives
> >> messages in its mailbox. And handle them one by one, just one in the same
> >> time.
> >> >
> >> > But in some situation, the request just want to read the resource. So
> >> the read requests should be able to execute at the same time since they
> >> don't change or break anything.
> >> >
> >> > So dose this meas if we use a write lock in such situation, we will
> >> have a better performance? If so, does Erlang have a proper way to do that?
> >> >
> >> > Thanks.
> >> >
> >> > ---
> >> > Bin Wang
> >> > _______________________________________________
> >> > erlang-questions mailing list
> >> > erlang-questions@REDACTED
> >> > http://erlang.org/mailman/listinfo/erlang-questions
> >>
> >>
> >
> > _______________________________________________
> > erlang-questions mailing list
> > erlang-questions@REDACTED
> > http://erlang.org/mailman/listinfo/erlang-questions
> >
> >
> 
> 
> -- 
> Regards,
> Xiao Jia

> _______________________________________________
> erlang-questions mailing list
> erlang-questions@REDACTED
> http://erlang.org/mailman/listinfo/erlang-questions