I have several PepWave MAX HD2 units with hardware revision 2 and revision 4. I am using an AT&T sim card and getting around 30mbps down and 2-4mbps up. I noticed when 2 or more people are connected to the MAX HD2 at the same time either wired or wireless (using the integrated AP), the bandwidth for each person is reduced greatly. For example, for 2 people connected at the same, each person is getting around 7-8mbps.
I am trying to understand how the 30mbps bandwidth is shared among the users connected at the same time. So, if I have 2 people connected to the MAX HD2 at the same and the bandwidth is 30mbps, would each person have 15mbps each? If I have 3 devices connected to the MAX HD2, would each device get 10mbps each?
Is this the way this technology works or is there a configuration setting in the MAX HD2 to allow each device 30mbps each assuming I’m getting 30mbps down from AT&T.
How are you measuring this? Are both users running a speedtest at the same time?
If all users generate traffic at the same time, the HD2 queues that traffic in its buffers as it arrives and its then sent on out through the active healthy WANS. The end result is that bandwidth will be shared.
If you have three devices and they all run speedtests at the same time, TCP’s congestion control mechanism will kick in the total bandwidth will be distributed across the LAN devices.
The accumulative bandwidth result from two devices running speedtest.net tests should equal (pretty much) the bandwidth result from a single Lan device running speedtest.net. there are no additional settings required on the hd2 to achieve this.
Another factor which can affect your total throughput and latency is Bufferbloat. The nature of TCP congestion is to build up the amount of data being transmitted, then drop back precipitously when congestion occurs, resulting in the sawtooth behavior seen in the great article Martin posted. The sawtooth behavior means that there are dips when your bandwidth is not being fully utilized.
Smart Queue Management (SQM) in the form of the fq_codel and Cake algorithms in routers deals with this, smoothing out the flow of data and ensuring fairness between different TCP connections. I’ve seen this make amazing improvements in download speeds when sawtoothing is avoided, and more importantly for most people, ensuring low latency. SQM requires upload and download maximum bandwidth to be configured somewhere between 5-10% less than the provisioned speeds, a price most people are willing to pay.
Peplink has partially enabled fq_codel on the Balance and Surf Soho routers as Mitigate Bufferbloat on their support.cgi page (on the home page of your router, change “index.cgi” in the url to “support.cgi”). It currently manages to the WAN upload value but is not managing the download side or even honoring the WAN download value as a maximum. Peplink hasn’t shared when they will fully implement fq_codel.
If the MAX HD2 has the Mitigate Bufferbloat feature, you may want to try it to see if it helps (I don’t have MAX experience, but Martin can probably comment). See this lengthy article for more details on Bufferbloat and Mitigate Bufferbloat:
Another feature that you can experiment with if it is available on the MAX is “DSL/Cable Optimization”. Martin has previously posted that it prioritizes acks. On a Balance it is under Network->QoS->Application. It is mutually exclusive with Mitigate Bufferbloat.