I’m doing lots of network calls over TCP to FEZ-Spider (FW version is 4.3.7.10).
I notice that often the client that communicates with the FEZ-Spider takes a lot of time to receive answers from the FEZ-Spider. I used Wireshark (=network sniffer) and I saw that probably the TCP packet with the answer was lost, and it takes the FEZ-Spider additional ~20 seconds to re-transmit the TPC packet.
this is a big problem
BTW: if the client loses a TCP packet, it re-transmit it in ~300 milliseconds!
How can I reduce that time?
If it is a FW issue, will it be solved in the coming GHI FW?
If it is a HW issue, is this issue solved in the other main boards / FEZ-Spider 2?
@ Matan - It is probably due to the known issues with the NETMF networking stack with regards to timers. Unfortunately that won’t be fixed until 4.4. Do you have other boards you can try it on?
@ John - tried it on FEZ-Spider 1.0. Now I tried it on FEZ-Spider 1.1, and it seems to have less re-transmissions (not sure) but it still take 20-30 seconds to do it.
@ John - I’m using the Ethernet J11D Module. Should I try also Ethernet ENC28 Module? is it more stable?
In the code I’m using the class
, and do Bind+Listen(queue-size=16)+Accept on a TCP socket.
By the way, I just got FEZ Spider II Mainboard that it is not EMX based (probably you know it is G120E based).
Is the timing better with this new mainboard?
@ Matan - We do not have information on the timing to be able to tell you if it is better. We have not noticed any issues beyond the known NETMF issues in our tests before. If you can, try with the ENC28 and the J11D and ENC28 on the Spider II so we can try to narrow things down.
