report browser and error_logger
Thu Jan 10 12:07:40 CET 2002
That is the exact reason why I did not suggest global. Martin's
application sounded like it was handling massive concurrency. Due to the
atomic implications of registering a process on all nodes, it sounds
like a rather expensive operation. Martin knew on what node the process
was running on, so using a local name server, data duplication and
global registering could be avoided (At the expense of having to handle
the operations himself).
The thought of tens of thousands of registered processes sends shivers
down my spine.... I'll pass on that :-)
Richard Carlsson wrote:
>On Thu, 10 Jan 2002, Ulf Wiger wrote:
>>I agree with Francesco. Another thought is to use global. While it
>>may not seem like an obvious choice, global does support process
>>registration using any Erlang term (instead of just atoms), and the
>>semantics of global:register_name() and global:send() are basically
>>the same as for your preferred approach.
>Just one thought: does anybody know how well the global name registry
>scales? Is it feasible to register thousands or tens of thousands of
>processes? How does it scale with the number of connected nodes?
>Richard Carlsson (richardc@REDACTED) (This space intentionally left blank.)
>E-mail: Richard.Carlsson@REDACTED WWW: http://www.csd.uu.se/~richardc/
More information about the erlang-questions