[R] socket clusters on snow dies easily

[Luke Tierney]

If you absolutely have to use the socket approach you will probably
have to track down the particular reason for this failure in your
environment by looking at what is going on under the hood and checking
the pieces.  The master sets up a server socket, the slave noted
connect to that socket, and for some reason the first read on that
connection seems to be failing.  It could be a timing issues--you may
need to adjust socket timeouts, it could be a permission or firewall
issue; hard to tell without experimenting.

You are probably better off using the PVM version if at all possible.
Overall it more solid than the socket version and likely to have fewer
issues.

Best,

luke

On Fri, 9 Apr 2004, Rong-En Fan wrote:

> hello,
> 
>   I'm using R 1.8.1 with the lastest snow package on FreeBSD 4.9.
>   However, when I try to using socket clusters, it's very unstable.
>   Sometimes it dies half way when I run parSapply(), sometimes
>   it dies when cluster connection is idle.
> 
>   I create a socket cluster by following cmd
> 
>   cl = makeCluster("foo", type = "SOCK", outfile="/tmp/rafanlog");
> 
>   Then, I just idle in R, and tail -f outfile, I got following:
> 
>   [... some R msgs ...]
> 
> Type 'demo()' for some demos, 'help()' for on-line help, or
> 'help.start()' for a HTML browser interface to help.
> Type 'q()' to quit R.
> 
> 
> >
> > #library(serialize)
> > library(snow)
> >
> > slaveLoop(makeSOCKmaster())
> Error in unserialize(node$con) : error reading from connection
> Execution halted
> 
> 
>   after makeCluster() returns, the outfile hangs around the empty lines
>   above, about 1 minute later, all msg shows up, also it dies.
> 
>   This is reproducible on my machine. Also, one of my friends also 
>   got this result on Linux machines w/ R 1.8.x.
> 
>   Is there any one also encounter this problem?
>   Any comments or suggestion are welcome :)
> 
> Regards,
> Rong-En Fan
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://www.stat.math.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html
> 

-- 
Luke Tierney
University of Iowa                  Phone:             319-335-3386
Department of Statistics and        Fax:               319-335-3017
   Actuarial Science
241 Schaeffer Hall                  email:      luke at stat.uiowa.edu
Iowa City, IA 52242                 WWW:  http://www.stat.uiowa.edu