[Swift-devel] coaster io with NIO.

Michael Wilde wilde at mcs.anl.gov
Mon Apr 9 10:48:39 CDT 2012


Mihael, this sounds very good. I replied with a comment on Bug 690 regarding fully debugging the root cause of the timeout problem.

But regarding the limitations between the new NIO code and GSI: the current secure socket solution for coasters has long been a usability obstacle. This sounds like a good time to design a highly usable solution that will last us into the future. What alternatives can you suggest for this?

- Mike

----- Original Message -----
> From: "Mihael Hategan" <hategan at mcs.anl.gov>
> To: "Swift Devel" <swift-devel at ci.uchicago.edu>
> Sent: Monday, April 9, 2012 10:06:56 AM
> Subject: [Swift-devel] coaster io with NIO.
> So I wrote this NIO version of the coaster send/multiplexer loops. It
> uses selectors instead of whatever was there before.
> 
> The multiplexer (i.e. reading) wasn't and issue because there is a way
> to test whether a read will block or not, but writing data to a
> collection of sockets without NIO suffers from one socket blocking
> everything because there is no way to test that a write on a socket
> will
> not block.
> 
> The initial implementation was not NIO based because there is no easy
> way to convince the jglobus GSI libraries to work with NIO, so my NIO
> thing only applies to plain TCP sockets. I should probably deal with
> that, although it's not an issue when you have a single coaster
> service.
> 
> I'll commit the changes today so that you guys can test it. Works fine
> on localhost so far.
> 
> Mihael
> 
> _______________________________________________
> Swift-devel mailing list
> Swift-devel at ci.uchicago.edu
> https://lists.ci.uchicago.edu/cgi-bin/mailman/listinfo/swift-devel

-- 
Michael Wilde
Computation Institute, University of Chicago
Mathematics and Computer Science Division
Argonne National Laboratory




More information about the Swift-devel mailing list