[zeromq-dev] Java/Windows issues

[Conrad Steenberg]

Hi Martin, 

On Tue, 2009-08-18 at 10:13 +0200, Martin Sustrik wrote:
> > The disadvantage to nesting the different functionalities is there is 
> > some duplication. TCP and PGM have their own less persistent session 
> > ids/seq numbers while the ZRecoverySock must have its own. Makes your 
> > message payloads slightly bigger. Doesn't really matter for file based 
> > stores be perf is so slow but a large in memory store would suffer. Of 
> > course a ZRecoverableSock could run over a UDP/IP socket and do pretty 
> > well (maybe needs windowing)
> 
> Additional sequence numbers on top of TCP/PGM are unaviodable for 
> guaranteed delivery (the one that survives component restart). As you 
> say, UDP is an option, however, it would mean reimplementing all the 
> nice TCP functionality (thing of congestion control mechanisms) so I 
> believe few additional bytes per message are worth of it.

My 2c on UDP:
In the long run UDP would of course scale better because it doesn't have
the kernel or networking overhead of TCP.

Just FYI, coming from an HPC/long,fat pipe networking environment, take
a look at whether the UDT library can be used instead of raw UDP/IP:
http://udt.sourceforge.net/ 

Cheers,
Conrad

> 
> Martin
> _______________________________________________
> zeromq-dev mailing list
> zeromq-dev at lists.zeromq.org
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev