Hi,Andy Thank you for your reply. Where is the "Documents/ppp-generic.txt" mentioned in your letter?I'd like to have look and find some clues. In TR059,it requires to implement such a mechanism as MLPPP LFI.I think the essence is LFI,which do fragmentation and interleaving.This method will reduce the delay of EF packets such as voip packets.So,in my viewpoint,two phases would be done on linux kernel or linux iproute2: 1.fragments the BE and AF packets if they are bigger than specific size. 2.put the following EF packets into the middle of these new fragmented packets. But It's so taugh for my current knowledge to implement these,especially phase 2.Even I can't find which place in linux kernel or iproute2 I would do these . What's your opinion?Any suggestion is highly appreciated! Thank you very much! ======= 2004-06-02 06:48:50 You wrote：======= >swcims wrote: >> Hi,Andy Thank your very much! For the MLPPP LFI,I found that in Cisco >> configuration,it use "ppp multilink;ppp multilink fragmentation;ppp >> multilink fragment-delay 20;ppp multilink interleave " command to >> enable MLPPP LFI.So I think just realizing the same function on my >> linux router would be fine.But I got no idea how to do this on >> linux.So is there anything with iproute2?Would you please supply some >> suggestions? As for "The packet size threshold before fragmenting AF >> and BE packets MUST be configurable", I am completely lost.Any advise >> is highly appreciated! Thank you again! > >I found the TR059 Doc - but haven't read properly yet. > >Grepping my sources - linux mlppp is mentioned in >Documents/ppp-generic.txt - it chooses frag size and expects > 1 line, >as I expect the Cisco does, unless it has other Qos settings the fact it >has settings for delay/frag etc. probably has nothing to do with TR059. > >In the Doc MLPPP is used as an example of how to fragment - more work is >needed to actually use it for QOS on one link. > >Andy. > > >> >> >> >> >> ======= 2004-05-30 15:29:20 You wrote：======= >> >> >>> swcims wrote: >>> >>>> Hi,Andy I don't understand what you mean.I have a adsl router >>>> based on mips linux2.4.17.In this adsl router,one side is 100Mbit >>>> ethernet card,another side is adsl line,as following: >>>> LAN------(ethernet)--------ADSL Router------(ADSL >>>> line)-------------WAN On this adsl line,it can configue >>>> rfc1483bridge or router,or pppoe to get wan ip from ISP. I think >>>> this side(upstream) is the bottleneck link. I ported iproute2/tc >>>> on this router and i can control traffice on the upstream line. >>>> But I got completely no idea for doing these: (from TR059 >>>> Technical Report DSL Forum) 1.The device MUST support the >>>> capability to fragment AF and BE traffic in order to constrain >>>> the perturbing impact of AF and BE packets on EF traffic delay, >>>> for example using a mechanism such as MLPPP LFI.(RFC1990) 2. The >>>> packet size threshold before fragmenting AF and BE packets MUST >>>> be configurable. Thank you very much! >>> >>> What I mean is that if you want to mess around with packets below >>> ip level, the other end - your ISP/teleco , will need to be >>> running software that knows what you are doing so that it can >>> reconstruct the packets before routing. >>> >>> If you have a specific need for your upstream not to be delayed >>> more than X ms you could adjust your MTUs/MSS clamp - the size will >>> depend on your bitrate and max delay required. >>> >>> Andy. >>> >>> >>> >> >> >> = = = = = = = = = = = = = = = = = = = = >> >> Regards. >> >> swcims swcims@xxxxxxx 2004-05-31 >> = = = = = = = = = = = = = = = = = = = = Regards. swcims swcims@xxxxxxx 2004-06-02 ???????????????????ē?%?,S ????醑?m??ē????????-?棹V?廖Y3???????
