Re: [Jack-Devel] Noisy Signal - Fragmented Data

On Wed, April 13, 2016 2:01 am, Kenneth Fields wrote: Independent things, a "good IPv6 network" does not necessarily imply jumbo frame support. The network stack would still need to determine path MTU before using large frames. I did however find a reference on the Internet2 web page that said the intention was to configure for 9kB frame support everywhere on Internet2. There are lots of places a link could be misconfigured, though. This link explains some of path MTU determination. It mentions some differences with IPv6 but does not discuss: http://packetlife.net/blog/2008/aug/18/path-mtu-discovery/ There are some references there and in comments left about how to discover MTU in a path and determine where MTU changes. The IPv6 specification section 4.5 has this note: (Note: unlike IPv4, fragmentation in IPv6 is performed only by source nodes, not by routers along a packet's delivery path -- see section 5.) Section 5 is titled "Packet Size Issues" and has this note: It is strongly recommended that IPv6 nodes implement Path MTU Discovery [RFC-1981], in order to discover and take advantage of path MTUs greater than 1280 octets. However, a minimal IPv6 implementation (e.g., in a boot ROM) may simply restrict itself to sending packets no larger than 1280 octets, and omit implementation of Path MTU Discovery. To me that sounds like the platform IPv6 layer should handle MTU discovery, but I'm not sure if application level software has to participate in that or not. I don't have any experience with the low level details of programming network software in C or C++, in higher level languages the libraries take care of a lot of that stuff behind the scenes, and for software that only runs on a LAN I just assume that 1500 byte packets are supported and plan for that. The page for jacktrip has this description: "It supports any number of channels (as many as the computer/network can handle) of bidirectional, high quality, uncompressed audio signal steaming." So in a case of 32 channels of 24 bit audio, that would be by your calculation: 32 x 3 x 128 = 12288 bytes 12288 bytes is the largest MTU I have seen used on switches, but that is raw Ethernet packet size, not including Ethernet headers, IP headers, and UDP headers. That means that 32 channels would not fit into a single frame of any equipment I have seen, much less the more typical case of 1518 byte Ethernet MTU, so it is unlikely that jacktrip would require that the entire audio frame fit into one Ethernet packet. I suppose it is possible, the documentation I found doesn't really say what is meant by "as many as the computerer/network can handle." My assumption would be that is payload bandwidth limited, not packet size limited. You might have to dig into the source to check. My first assumption would be that jacktrip has been around long enough it would have come up before now if packet size was really a problem. Did you actually use the exact same configuration before? Same end point locations, same network route? If you switched from IPv4 to IPv6 then the addresses in the header went from 32 bit addresses to 128 bit addresses. IPv4 headers are 20 bytes and IPv6 headers are 40 bytes, so you lose 20 bytes of payload if you switch from IPv4 to IPv6 and keep the same Ethernet MTU. Even if you were using IPv6 on your comparison, something could have changed in the path. You would need to compare traceroute results from previous connections and current connections to see if the routing is the same or has changed. The route could have changed, or even on the same route equipment could have changed, or even just a configuration mistake of the same equipment on the same route. And of course packet fragmentation could be unrelated to the problems you have, presumably you didn't analyze the transmission with wireshark when everything was working properly, so you don't know if the audio data to packet distribution is the same or different. You could compare to a working connection and see if the behavior is really different. If for some reason the jacktrip application is handing down large packets when the path does not support packets of that size, it would be considered bad design practice. I think (not sure) that the IP layer may fragment for you, but RFC 5405 says you should not do that: "Due to these issues, an application SHOULD NOT send UDP datagrams that result in IP packets that exceed the MTU of the path to the destination. Consequently, an application SHOULD either use the path MTU information provided by the IP layer or implement path MTU discovery itself [RFC1191][RFC1981][RFC4821] to determine whether the path to a destination will support its desired message size without fragmentation." Are you sure the problem isn't just plain old dropped packets? Does jacktrip give an indication if a sequence number is missed? Presumably only the receiving side would know since UDP does not send ACK packets. The only design document I could find was here: https://ccrma.stanford.edu/groups/soundwire/publications/papers/2009-cacere… That document explicitly says that clocks are not synchronized, which means you should EXPECT to hear glitches regularly: "...and deals with clock drifts and late or un-recoverable packets by using a lower-level strategies in ring buffers that can, e.g., sound like a wavetable synthesizer, thus extending the musical sonority of the moment. Clock drift between remote WAN machines is still an unsolved issue and there are presumably new techniques to be tried in the future, like adaptive re-sampling, packet crossfading, and others." Sounds like jacktrip is a basically broken design for network audio, and you are going to get a continuous stream of glitches by design. -- Chris Caudle