toppin@melpar.UUCP (Doug Toppin) (02/23/89)
For TCP/IP: In our environment we have to pass a large number of small messages from a number of processors to a single processor. Due to max open socket limitations on the server the connections cannot be kept open for more than a single message. We are using IBM Xenix V on the AT with NRC Fusion TCP/IP. I am interested in using data grams but the unreliability problems in lost or duplicated messages are a concern. My Questions: * Is there a performance benefit in using the SOCK_SEQPACKET over the SOCK_STREAM options in the socket call (given known data sizes to be sent)? * How unreliable are data grams in a system without a gateway in the real world? (percentage of error) * What is the best approach in sending a large number of small messages where no losses or duplicates getting to the applications software can tolerated? * Is there software in the public domain that will provide a reliable data gram system and fit between applications software and the protocol software? Thanks in advance, reply via mail, useful replies will be posted. Doug Toppin uunet!melpar!toppin
braden@VENERA.ISI.EDU (02/24/89)
Doug, WHAT YOU NEED IS... a tranaction transport protocol. I suggest you look into Dave Cheriton's VMTP, described in RFC-1045 (see also RFC-955). Bob Braden