There should be no MT router in use today that has performance issues with single customer standard NAT. Maybe 10mbit+ plans on 10+ year old routers, but there shouldn't be a difference.
That said, Fast Track does greatly improve the performance of anything using connection tracking. ----- Mike Hammett Intelligent Computing Solutions http://www.ics-il.com Midwest Internet Exchange http://www.midwest-ix.com ----- Original Message ----- From: "Paul McCall" <pa...@pdmnet.net> To: af@afmug.com Sent: Wednesday, July 22, 2015 5:59:38 AM Subject: Re: [AFMUG] Epmp user throughput I posed about our similar frustrations a while back… here’s an update. I would suspect something in the router’s queuing methods as well. Just a big important FYI, for those of you using Mikrotik routers and NAT for customers, if you want customers to always be able to achieve full speeds, use 6.29 or above an enable FastRack. We had a very difficult time, only with some customers getting full speeds. 2 computers could get each get 2.8 Mbit at the exact same time, but no individual PC could get more than 3Mbit. Not all users had this and the weak response “from the crowd” was the wireless link (even some good links from AP to SM) was causing a little extra queuing at the hardware level, resulting in performance behind the router to be less than optimal. Yeah, I know is not a completely scientific description that we can all latch onto, but it appears to have validity. Here are the FW rules for that version that magically fixed the few customers that we have tried it on. /ip firewall filter add chain=forward action=fasttrack-connection connection-state=established,related add chain=forward action=accept connection-state=established,related add chain=forward action=drop connection-state=invalid As the name says, it Fast tracks most of the firewall (processing of queues and other rules) which may not work for all your situations, such as a heavy user that absolutely needs VoIP optimization because they are slamming their connection. (not the norm) Paul From: Af [mailto:af-boun...@afmug.com] On Behalf Of Ken Hohhof Sent: Wednesday, July 22, 2015 6:44 AM To: af@afmug.com Subject: Re: [AFMUG] Epmp user throughput Also, the two tests may be different. Speedtest.net type test may be saying I see packet loss starting at X Mbps so the rate is X, while iPerf type test may say I see Y Mbps goodput despite some packet loss so the rate is Y. This may be interacting with the queuing method, as Mike points out. Perhaps set the ePMP to wide open and see if customer results change, then set queue in Miktrotik and see what happens. From: Mike Hammett Sent: Wednesday, July 22, 2015 5:30 AM To: af@afmug.com Subject: Re: [AFMUG] Epmp user throughput I read it as using the ePMP for queuing, the performance is less than expected. Using the MT for queuing, the performance is as expected. ----- Mike Hammett Intelligent Computing Solutions http://www.ics-il.com Midwest Internet Exchange http://www.midwest-ix.com ----- Original Message ----- From: "Josh Luthman" < j...@imaginenetworksllc.com > To: af@afmug.com Sent: Tuesday, July 21, 2015 10:57:18 PM Subject: Re: [AFMUG] Epmp user throughput If the Mikrotiks at the customer site are doing 40x10 but the customer devices behind the MT it doesn't really make sense to look at the epmp for your problem. It could be the Mikrotik's CPU, port, wireless or the customer device. Josh Luthman Office: 937-552-2340 Direct: 937-552-2343 1100 Wayne St Suite 1337 Troy, OH 45373 On Tue, Jul 21, 2015 at 11:54 PM, Nate Burke < n...@blastcomm.com > wrote: I have deployed quite a bit of epmp, but most people are at my base package, 4x1 (using the radio qos to limit bw on different tiers). Of the handful of people that are on larger plans, 20x5 or 40x10, 2 of them, on different towers are complaining that they struggle to get over 10mb on a speed test. The radio rf link test performs at the assigned qos level, and I sent one of them a mikrotik to go between the poe and his router (qos 40x10), and udp/tcp tests to that are coming back as expected. However, I can be watching the router, and his interface plugged directly into a laptop only runs like 10x2 during a speedtest. Do both of these customers have something wonky with their laptops, or is there a setting in epmp I'm overlooking? Most of our epmp deployment has been FSK upgrades, so I haven't had much actual laptop time myself behind an epmp sm, as we just go on the roof and change radios. Nate
