That's not normal behavior at all ... how are you measuring that LLNW (or provider X - doesn't matter who) is sending greater than the amount of traffic the customer can handle? Most CDN's (and LLNW isn't bad at this - not super great) have numerous adaptive bitrate "profiles" for delivery so ensure they are using what's available ... it's not in their best interest to completely flood the link neither ..
-----Original Message----- From: Af [mailto:af-boun...@afmug.com] On Behalf Of Ken Hohhof Sent: Wednesday, July 8, 2015 10:26 PM To: af@afmug.com Subject: Re: [AFMUG] Limelight Networks and TCP congestion control - 1000 Let me clarify. Customer has subscribed to a certain speed tier, let's say 5 Mbps. So LLNW consistently sends 10 Mbps of traffic to this customer, half of which gets dumped on the floor. Actually half of the customer's traffic gets randomly dropped, not just the LLNW traffic. The #1 problem with this is that one edge provider is inflicting 50% packet loss on every other edge provider trying to send content to this end customer. Everything else the customer tried to do, even loading simple web pages, was dog slow. Not because his Internet connection was "slow" or because his ISP didn't have enough bandwidth. But because LLNW was crowding every other provider out. I don't see how this practice by LLNW is efficient delivery, or how you interpret this to mean the ISP does not have enough bandwidth. Unless you interpret selling different speeds at different prices as not having enough bandwidth. -----Original Message----- From: Paul Stewart Sent: Wednesday, July 08, 2015 7:25 PM To: af@afmug.com Subject: Re: [AFMUG] Limelight Networks and TCP congestion control Yeah ... just think of "whack a mole" here.. you slow down CDN network X .... this may force the traffic (depending on who is involved here) to come from a different node or even a completely different CDN network .. not sure why LLNW are considered dicks here - their job is to deliver the content as fast and as efficiently as possible ... ISP job is to have enough bandwidth to deliver if possible (I know, loaded statements).... Paul -----Original Message----- From: Af [mailto:af-boun...@afmug.com] On Behalf Of George Skorup Sent: Wednesday, July 8, 2015 8:15 PM To: af@afmug.com Subject: Re: [AFMUG] Limelight Networks and TCP congestion control Is this only from LLNW? I swear I've seen it from other CDNs as well. I don't see any way around this other than throwing bandwidth at it. And as you mention, this means upstream bandwidth and backhaul links between towers. On 7/8/2015 5:50 PM, Ken Hohhof wrote: > This has been discussed in previous threads, but I just got off a call > with a customer where I was able to identify what content was being > distributed via LLNW and causing problems. Customer had bought a new > Xbox and it was downloading game updates. He was complaining that the > game update seemed stuck getting to 100%, plus Internet on his other > devices was painfully slow. > > LLNW seems to aim for 50% packet loss. I always seem to see almost > exactly 2X the customer's rate limit hitting our border router. Of > course this causes all traffic to that customer to experience random > 50% packet loss, which few applications can tolerate. It is also > consuming our upstream and backhaul bandwidth beyond what the customer > has subscribed to, since we do rate limiting at the tower router. > > Other than getting a Procera box and setting up some kind of rule to > catch this at the border (plus maybe a penalty for LLNW being such > dicks), I don't know what to do about this, but it pisses me off. >
