Cambium answer of this Problem: Say you have two routers, router 1 at 10/100 for port 1 and 10/100/1000 for port 2 and router 2 communicates with router 1 using the 10/100/1000 and only has 10/100/1000 ports. Router 2 receives a burst of traffic and sends it to router 1 over 10/100/1000 port. Router 1 gets the traffic, but cannot send it out as quickly over its 10/100 port. Router 1 starts buffering the data. But routers do not have very big internal queues so some data gets dropped. ePMP because it has frames, send data in bursts. In the direction the data is sent, it can accumulate lots of data each frame. In the direction of the TCP ACKs, it can accumlate lots of these every frame. For the case of the data, this data can be lost at router 1 if it is after the ePMP device. For the case of the TCP ACKs, router 1 can handle the TCP ACKs load, but when the device that initiated the TCP session receives the TCP ACKs at once, it responds with new TCP data all at once and this causes a burst. If router 1 receives this burst of data, then data gets dropped. Whenever TCP data is dropped, this causes the TCP window to shrink which slows down throughput. Therefore, if you have the situation I described, you need to change your routers so they all have the same port speed.
2016-03-11 23:08 GMT+01:00 Josh Luthman <[email protected]>: > Handful of orders these last couple weeks. 0 differences beyond the Ubnt > out of stock stuff (yes that record is still broken). > > > Josh Luthman > Office: 937-552-2340 > Direct: 937-552-2343 > 1100 Wayne St > Suite 1337 > Troy, OH 45373 > > On Fri, Mar 11, 2016 at 5:06 PM, Ken Hohhof <[email protected]> wrote: >> >> I think they were doing end of fiscal year inventory the last couple days. >> >> From: Bill Prince >> Sent: Friday, March 11, 2016 3:07 PM >> To: [email protected] >> Subject: Re: [AFMUG] Poor throughput on ePMP AP >> >> Speaking of Streakwave. Is everyone else having issues with the deliveries >> getting stretched out beyond what seems normal with them? >> >> bp >> <part15sbs{at}gmail{dot}com> >> >> On 3/11/2016 1:01 PM, Kurt Fankhauser wrote: >> >> i can't get a 3011 to try, streakwave doesn't have stock, but i >> desparately need this, have a tower with high CPU usage on router and only a >> 2011 fill fit in the enclosure >> >> On Fri, Mar 11, 2016 at 3:59 PM, Bill Prince <[email protected]> wrote: >>> >>> Funny. We were just talking about this yesterday. The 2011 is great for >>> small POPs, except for the split 10/100 - 1000 ports. The 3011 would be >>> perfect except it can only be had in a 19" rack form factor. >>> >>> Has anyone tried to stick the circuit board from a 3011 into a 2011 case? >>> DC plant of course. >>> >>> bp >>> <part15sbs{at}gmail{dot}com> >>> >>> On 3/11/2016 12:43 PM, Ken Hohhof wrote: >>>> >>>> My rackmount 2011's will probably get replaced with 3011's, but they >>>> really need to bring out a smaller desktop 3011 as well. >>>> >>>> >>>> -----Original Message----- From: Adam Moffett >>>> Sent: Friday, March 11, 2016 2:23 PM >>>> To: [email protected] >>>> Subject: Re: [AFMUG] Poor throughput on ePMP AP >>>> >>>> Well that's news to me! Holy crap. They should put a warning sticker on >>>> that. >>>> >>>> On 3/11/2016 2:20 PM, Joe Falaschi wrote: >>>>> >>>>> The other thing to keep in mind with the RB2011 is what port things are >>>>> plugged into. Ports 6-10 only have a 100M aggregate link to ports 1-5. If >>>>> the aggregate of ports 6-10 require more than 100M, you'll have issues >>>>> there >>>>> too. >>>>> >>>> >>>> >>> >> >> >> >
