[erlang-questions] 500k user simulation on EC2

[Witold Baryluk]

Dnia 2009-08-13, czw o godzinie 20:05 +0100, Joel Reymont pisze:
> I'm creating a simulation on EC2 where I need to launch 500k clients.  
> The clients need to connect to a cluster of servers, exchange packets,  
> report latency and exit.
> 
> I need to report the simulation as unsuccessful if a client fails and  
> otherwise wait for all my clients to exit. I figured that I can make  
> clients part of a simple_one_for_one supervisor. This way I can ensure  
> that the simulation fails if at least one client fails.
> 
> Would a supervisor have trouble with 500k pids on different nodes,  
> though?
Very big trouble. generic supervison have pid in list, just
finding pid in list, and removing it will be disaster I think,
and it will be gigantic and with lots of redundancy (same module
names there). Try testing, and maybe change it to fast data structure.

I would suggest taking OTP's supervisor source, and modifing it.
And benchmarking it in the same time. Keep it simple and working.




-- 
Witold Baryluk <baryluk@REDACTED>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 197 bytes
Desc: To jest cz??? wiadomo?ci podpisana cyfrowo
URL: <http://erlang.org/pipermail/erlang-questions/attachments/20090813/3e9c7f02/attachment.bin>