Newsgroups: comp.os.parix
From: asterix@ciunix.uc.pt (Patricio Rodrigues)
Subject: 'Select()' overhead ?
Organization: Dep. de Matematica da Univ. de Coimbra
Date: 24 Oct 1995 15:54:33 GMT
Message-ID: <46j27p$pnf@gnu.mat.uc.pt>

	Hi,

I've been using the 'Select ()' ( SELECT(3x) ) in a PowerXplorer machine.
I'm using the 'alternative wait' option, i.e. 'ReceiveOption(...)' 
for configuration and 'SelectList()' to wait for the next message. 
Everything works fine but rather slowly...

The communication pattern of my problem can be decribed as follows:

A master waits for the results of N (actually 7) workers. To avoid a 
serialisation in the receive operations which would happen using 
simply RecvLink() ( receive from worker 1, the receive from worker 2,
 ..., receive from worker N), I use the Select() feature so 
communications can be made in a freer manner, only calling 
'RecvLink( link )' when a message is effectively available on 
that link.

So, I have the same questions:

a ) Is there anyone with similar complain about 'Select()' 
performance overhead ?

b )  Does anyone know how 'Select' is implemented at internal level ?

	Thanks in Advance,
	Patricio.

P.S.: Please e-mail the answers since my news server is frequently 
down. Thanks!


