dnwcv@dcatla.UUCP (William C. VerSteeg) (06/10/88)
I recently spoke with a vendor of host based TCP/IP software. I was trying to determine the company's support of ICMP Source Quench. The package they sell ignores ICMP Source Quench messages. It doesn't lower offerred traffic when receiving a source quench and doesn't send a source quench when its buffer space is running out. I requested a fix for this problem. We discussed some of the ongoing work in the field of congestion control and its impact on the usability of source quench mechanisms. He informed me that the company had no plans to implement source quench support due to the large number of implementations that also ignore source quench. This brings me to my point. How many of the currently popular packages handle Source Quench in a reasonable fashion? Is it worth my efforts to chase this vendor and force him to conform to specifications or is it a mute point due to a general lack of conformance to the ICMP spec by most of the available packages? We definately require a traffic limiting mechanism. I don't know if the ongoing congestion control work being done obviates the need for source quench support, but I doubt that it does. Bill VerSteeg
Mills@UDEL.EDU (06/11/88)
Bill, At least the new Jacobson/Karels TCP does listen to source-quench and the present NSFNET Backbone system does generate them. According to good-faith info, the new NSFNET Backbone should generate source-quench in a repsonsible way within the next few months. I would suggest you beat on your vendor with this message. I should point out the present system is a far way from optimally tuned and that ongoing research here at UDel shows this tuning may not be easy; however, the handles are there and our experience shows they do work. Dave