[Spread-users] Optimal message size.

Yair Amir yairamir at cnds.jhu.edu
Thu Aug 9 20:08:35 EDT 2001


Well - not 1MB. 

Unless you changed the Spread code, 100KBytes or so is the largest possible message.
BTW - the exact number of bytes WILL change if you have multiple destination groups. Spread
will return an error if you try to send a message larger than the maximal message it can handle.

If you did change the Spread code to allow for 1MB messages (quite simple change), be aware that
THIS does have implications on memory footprint usually on the receiving Spread especially if you send many such messages and your receiver applications do not read them promptly. 


	:) Yair.
"Theo E. Schlossnagle" wrote:
> Yair Amir wrote:
> > Perhaps I did not understand your question. But if your messages are one order of magnitude larger than the Ethernet MTU (i.e. they are 5-10 KBytes) the best is to not split them and give them to Spread using one SP_multicast.
> >
> > Actually, for any size up to the max spread message, the best is not to split. The reason is that if you do not split, Spread optimizes its packets. If you do split unnecesarily, be aware that Spread will add a message header (pretty big - 80 - 100 bytes or so depending on the version and wide area support) to every "split" it gets.
> >
> > There is no relation between message size and memory footprint if the messages are large (5-10Kbytes). The relation is between (more or less) the total number of bytes that are pending (a different way to say - total number of spread packets).
> I think you understand the message.  You answered my question!  There is an
> overhead (albeit small) for each message I send.. And Spread optimizes to fit
> in the MTU of the media you are running on (or at least for Ethernet), so I
> gain quite a bit by passing a 1MB message to SP_multicast (or as big as it
> will take).
> Thanks!
> --
> Theo Schlossnagle
> 1024D/82844984/95FD 30F1 489E 4613 F22E  491A 7E88 364C 8284 4984
> 2047R/33131B65/71 F7 95 64 49 76 5D BA  3D 90 B9 9F BE 27 24 E7

More information about the Spread-users mailing list