Arnold, I configured a 4 container experiment and ran a quick test:
[Host 1] <- lan -> [Radio 1] <- OTA -> [Radio 2] <- lan -> [Host 2] LXC 1 LXC 2 LXC 3 LXC 4 iperf was run on Host 1 and Host 2 containers (UDP, 1400 byte packets, 100Mbps). I observed a relatively consistent 95Mbps without assigning specific CPUs to each container. Additional testing, which I did not perform, might include specifically assigning 1 or more CPUs to each container. The RF Pipe downstream queue is not controlled via a configuration parameter. If you wish to increase the queue size you'll need to make the change here: https://github.com/adjacentlink/emane/blob/master/src/models/mac/rfpipe/downstreamqueue.cc#L39 -- Steven Galgano Adjacent Link LLC www.adjacentlink.com On 07/27/2018 03:18 PM, Zhongren Cao wrote: > Hi Steve, > > We are interested in emulation a high speed wireless network using RFpipe > NEMs. To start, we setup a small example, in which only two containers are > active representing two network nodes. The “datarate” in rfpipemaclayer is > set to be 100M. After starting EMANE, we ran iperf udp to measure the > throughput between the two emulated radio nodes. The path loss is set such > that we should get zero packet loss. Thus, we expect to get a throughput > measurement very close to 100Mbps. However, we couldn’t. Instead 100 Mbps, we > could only get to about 83 Mbps. > > Upon investigation, we noticed that the transmitter’s RFpipe MAC dropped many > packets, as shown in the following statistics. > > nem 2 mac numDownstreamPacketsUnicastRx0 = 142669 > nem 2 mac numDownstreamPacketsUnicastTx0 = 105646 > > and > > nem 2 mac UnicastPacketDropTable0 > | NEM | SINR | Reg Id | Dst MAC | Queue Overflow | Bad Control | Bad Spectrum > Query | Flow Control | > | 1 | 0 | 0 | 0 | 37023 | 0 | 0 > | 0 | > > > The TX mac got 142669 unicast packets from its virtual transport layer bit > only sent 105646 unicast down to its phy layer. We also verified that these > 105646 packets was successfully delivered all the way to the RX virtual > transport and into the iperf server running at the RX node. > > How can we increase the buffer size to resolve the queue overflow issue? > > Thanks, > Arnold > > > _______________________________________________ emane-users mailing list [email protected] https://publists.nrl.navy.mil/mailman/listinfo/emane-users
