[HECnet] native Dup sync line revisited --> preliminary tests reveals problems

Err.. Well I stand corrected.? The logic I was referring to was for generic pack transmission through TMXR connected lines.? The APIs which provide that packet delivery functionality aren?t used for DDCMP connected circuits (I forgot this since it was 8 or 9 years ago). Instead, as you said, the packet data itself contains the needed length information.? The APIs I was referring to are tmxr_get_packet_ln_ex()and tmxr_put_packet_ln_ex().? The DDCMP circuits have framing logic code located in pdp11_ddcmp.h and they use ddcmp_tmxr_get_packet_ln() and ddcmp_tmxr_put_packet_ln().? Another protocol which had imbedded framing logic and packet sizes could implement its own {protocol}.h which handled that particular protocol framing and length details and yet still use the TMXR layer for connectivity and transport. *From:* owner-hecnet at Update.UU.SE <owner-hecnet at Update.UU.SE> *On Behalf Of *Paul Koning *Sent:* Sunday, December 12, 2021 1:27 PM *To:* hecnet at update.uu.se *Subject:* Re: [HECnet] native Dup sync line revisited --> preliminary tests reveals problems So... if DUP-only mode using TMXR framed transmission which throws in a size field on TCP, that would imply UDP mode should be used instead. ?Right? paul On Dec 12, 2021, at 1:46 PM, Mark Pizzolato - Info Comm <Mark at infocomm.com <mailto:Mark at infocomm.com>> wrote: The byte count is absolutely needed in TCP mode since the byte count describes the packet boundaries on the wire.? UDP mode is already packets hence the insertion of the packet length preceding the data so the next layer doesn?t (in the device simulation) doesn?t have to be aware of TCP or UDP. *From:*owner-hecnet at Update.UU.SE <mailto:owner-hecnet at Update.UU.SE> <owner-hecnet at Update.UU.SE <mailto:owner-hecnet at Update.UU.SE>>*On Behalf Of*Paul Koning *Sent:*Sunday, December 12, 2021 10:42 AM *To:*hecnet at update.uu.se <mailto:hecnet at update.uu.se> *Subject:*Re: [HECnet] native Dup sync line revisited --> preliminary tests reveals problems You mean DUP-alone? ?KMC-DUP is compatible with DMC, right? ?And that sends (in TCP mode) the frames as-is, no byte count since none is needed. paul On Dec 12, 2021, at 1:37 PM, Mark Pizzolato - Info Comm <Mark at infocomm.com <mailto:Mark at infocomm.com>> wrote: The functioning DUP simulation only sends packets.? If the connection is TCP, then a 2 byte packet size is stuffed ahead of the packet data on transmit and the receiving side strips this from the data presented to the simulated system.? In UDP mode, the UDP packet size is stuffed into the beginning of the packet buffer and that size is then also stripped just like the TCP case. The problems that Reindert is seeing are probably related to timing of byte delivery into the simulation or timing of successive packets being delivered which may possibly overrun buffer space in the simulated OS. *From:*owner-hecnet at Update.UU.SE <mailto:owner-hecnet at Update.UU.SE><owner-hecnet at Update.UU.SE <mailto:owner-hecnet at Update.UU.SE>>*On Behalf Of*Paul Koning *Sent:*Sunday, December 12, 2021 10:29 AM *To:*hecnet at update.uu.se <mailto:hecnet at update.uu.se> *Subject:*Re: [HECnet] native Dup sync line revisited --> preliminary tests reveals problems On Dec 11, 2021, at 7:41 PM, R. Voorhorst <R.Voorhorst at swabhawat.com <mailto:R.Voorhorst at swabhawat.com>> wrote: L.S. This is the follow-up from the Dup-without-Kdp in character mode discussion somewhat earlier, where I stated it was not supported per the specifications in Simh comments and Mark P had some comments about that it was supported. During a moment of spare time, I reactivated a triple node Rsx-11M+ test set, used for specific Decnet testing for Phase-I to IV. These triple nodes were setup for using every possible device for Decnet communications, amongst them the native (non-kdp) Dup. This one in multipoint mode is character based and is driven through a Rsx Ddcmp software driver to participate in Decnet communications. If Simh Dup simulation is supported in native character mode, this would establish sufficient proof for a well-functioning character based synchronous line. Yes, if you use TCP mode. ?It's not likely to work in UDP mode because then the DMC emulation expects to get a full DDCMP frame in a single UDP packet. ?But in TCP mode it just picks apart the byte stream. When the Dup line is interfaced to a Dmc line, the test reveals flapping line behaviour: line up -? circuit fault ? line down; when trying to transfer data an immediate circuit fault appears. When the Dup is interfaced to another Dup, there is packet exchange so basically it should be able to work. However, although there are no problems with the lower length (8) signaling packets, the moment 22 char packets are transferred, the receiver complains about bad header as well as bad data crc checksums, so the packets are somewhat mangled. ?It prohibits to start Decnet communications. It looks like this in the snippet below: DBG(10561953465)> DUP RCV: Line:0 0000 81 0C C0 00 01 01 1D DF 01 28 04 01 40 02 02 00 .........(..@ <mailto:..@>... DBG(10561953465)> DUP RCV: Line:0 0010 00 0F 00 00 1D 46?????????????????????????????? .....F DBG(10561953465)> DUP RCV:? rxnexttime=10561870955 (-20000 usecs) DBG(10561953465)> DUP PKT: Line0: <<< RCV Packet? len: 22 DBG(10561953465)> DUP PKT:*Data Message, Count: 12, Num: 1, Flags: SQ, Resp: 0,_HDRCRC: BAD, DATACRC: BAD_* ... Per Mark P?s comment that some kind of filtering takes place in non-kdp dup mode, that might indicate that the source of the problems could be located in that corner. That behaviour needs to be examined. I wonder what kind of filtering that might be. ?The header ?looks completely normal for a DDCMP data frame, it indicates a data length of 12 bytes which matches what you see in that packet. ?12 would be the expected length of a Routing Init message, normally the first data packet sent after DDCMP startup is complete. ? Then again, its contents make no sense as an Init message. ? Sending node 1.40, does that seem right? ?But the "tiiinfo" (node type etc.) is 0x40 which isn't valid. ?Nor is the routing version (0.0.15) or the hello timer value (0). Could it be that one side is set in TELNET mode while the other is not? ?The settings must match or 0xff bytes are mishandled. paul