ipc - Erlang Documentation/SMP: single-node and multi-node per machine or per application, and the confusion that may follow -
I am currently studying the process of Errang. I have hit a jerk at Arlong (section 3, paragraph 2):
It explains why in some cases why to run several SMP VMs with a scheduler and Instead of many schedulers, on a SMP VM with many schedules, of course, many VMs need to be run that the application can be run in many parallel functions, with no less or less communication with each other it happens.
Now this paragraph is misleading me; I can see the Uni-process multiple scheduler scenario, but I fail to see many processes with a scheduler; Probably each process will have a separate node name , and this means that without modifying a particular application, this model can not be used with; Due to the lack of modification is not required, is mentioned as a key feature of SMP in the report. If multiple processes have the same node name, then due to inter-arcing-process messaging storms performance - it considers the use of in-memory amnesia Is there some process model in the article Not presented and I am missing here?
The author says what is saying here? Is he trying to suggest that an application for multiprocessing solo scheduler case has to be rewritten (to take many unique node names in the account)?
- 1 Edit: Explanation of the source of the problem -
The question has been answered through discussion; In the following I have a framework of trouble.
The issue of this question is that the document, as I recall, does not touch the scenario of running several Erlang emulators for the physical machine - it has always been shown that the emulator is your physical machine ( In industrial use); Also, the scenario of dividing a program explicitly for computational efficiency has never been considered. This sudden introduction has been the source of my misery.
The conference is still biased towards creating lots of processes and in the future there have been many improvements for SMP emulator for Ehrling, and that means copy Assuming the machine single node is still a very viable option considering adopting favorable options.
After writing the article, rewrite:
That's why some SMP VM with many SMP VM can be much more efficient for running many SMP VM with several schedulers on why.
- Lock-lock in non-SMP VM is so fast
- Single scheduler SMP VM due to the cost of 10% slower, probe locks
- Again, due to the waiting / waiting for multiple scheduler SMP VM lock
Certainly many VMs require that the application perform many parallel functions Can be run in which there is no short or very little communication with each other.
- I think : Nodes have different names on the same server.
- Inter Process Messaging During Slow Motion of VM Node Due to Interstitial Prologue Chalk Interval Process Messaging
Comments
Post a Comment