Re: [OpenWrt-Devel] trying to acheive close to 1 Gbps with 802.11ac 3x3 mimo ath10k Compex wle900vx
Just to close off this discussion - I did some final tests on two different board to show the problem is clearly the CPU not keeping up All tests performed with 3 simultaneous TCP streams using iperf, 80MHz channel width, channel 48, and antennas arranged at 120 degrees, all systems running 18.06 openwrt Compex WPJ563 (Qualcomm Atheros QCA9563 74Kc MIPS processor, 64KB I-Cache and 32 KB D-Cache, targeted to operate at 775MHz, 128MB ram, 16MB NOR flash), with Compex wle900vx card * Internal iperf test to localhost: Max speed = 525Mbps * Apple Mac air (with 802.11ac 3x3 MIMO) <-> compex - Apple -> Compex: 305 Mbps - Compex -> Apple: 125 Mbps - CPU during test (56% used by iperf, 41% by ksoftirqd) .. its suffering Espressobin (dual core ARM Cortex A53 processor up to 1.2GHz, 1GB ram, ..Marvell Armada 3700LP) with Compex wle900vx card * Internal iperf test on localhost: Max speed - 3.49 Gbps * Apple Mac air (with 802.11ac 3x3 MIMO) <-> Espressobin - Apple -> Espressobin: 515 Mbps - Espressobin -> Apple: 459 Mbps - CPU during test ( 19% used by iperf -s -V, 12% by ksoftirqd) ... not breaking a sweat Raw 802.11ac 1.3Gbps (max TCP throughput of approx. 650 Mbps) ... so the Espressobin is quite close Cheers David On Sat, Oct 6, 2018 at 12:30 AM David Johnson wrote: > > It looks like Ben was right > > Further investigation is showing that my Compex WPJ563 (Qualcomm > Atheros QCA9563 74Kc MIPS processor, Dragonfly Series 64KB I-Cache and > 32 KB D-Cache, targeted to operate at 775MHz 128MB System Memory, 16MB > NOR Flash) is the bottleneck. > > I've tested on the 1Gbps Ethernet link between a laptop with a 1Gbps > port and the Compex 1Gbps port and the best throughput I got was 98 > Mbps. This was with 3 TCP streams on iperf. > > I'm still finding it hard to believe that I can't get more speed out > of this compex board running openwrt ... top is not showing the CPU go > over 30% with iperf when the test is running > > Thanks for the other advice - I've built a test rig with antennas > oriented like a propeller at 120 degrees (I'll share a pic) ... once > I sort out the underlying problem or upgrade my router board - I'll > test this mimo rig. > > David > > On Fri, Oct 5, 2018 at 6:02 PM Ben Greear wrote: > > > > If you are running iperf on a weak-ish AP CPU, then that will be a > > bottleneck. > > Run iperf on external system through the AP instead. For TCP, use multiple > > streams, > > and first test with Ethernet mode to make sure iperf and/or CPU is not the > > bottleneck > > before you move to testing the WiFi interfaces. > > > > We can get around 900Mbps UDP download throughput on a 'perfect' setup with > > 3x3 9880 WLE900VX > > systems and the ath10k-ct firmware. > > > > Forcing MCS rate is not a good idea...tune antennas and/or RF environment > > so that you do not have > > to force MCS to get highest encoding rate. > > > > Thanks, > > Ben > > > > On 10/05/2018 04:17 AM, David Johnson wrote: > > > I'm trying to see what max speed can be achieved > > > > > > My setup is as follow: > > > Board: Compex WPJ 563 > > > Card Compex wle900vx > > > OS: Libremesh running openwrt 18.06 > > > Driver: ath10k_pci :00:00.0: firmware ver 10.2.4-1.0-00033 api 5 > > > features no-p2p,raw-mode,mfp,allows-mesh-bcast crc32 c41417d0 > > > Firmware: firmware-5.bin_10.2.4-1.0-00041 > > > > > > Physical setup > > > Connected 3 6dBi antennas on each side to CH0, CH1, CH2 port to > > > achieve 3x3 mimo. Distance between 2 devices is 2m on a bench. Set > > > power level on each end to 8 dBm to ensure I don't saturate radios > > > > > > Status of radio Node A > > > iwinfo wlan0-mesh info > > > wlan0-mesh ESSID: "speedtest" > > > Access Point: 04:F0:21:3E:DA:A2 > > > Mode: Client Channel: 48 (5.240 GHz) > > > Tx-Power: 8 dBm Link Quality: 70/70 > > > Signal: -33 dBm Noise: -102 dBm > > > Bit Rate: 6.0 MBit/s > > > Encryption: none > > > Type: nl80211 HW Mode(s): 802.11bgnac > > > Hardware: 168C:003C : [Qualcomm Atheros QCA9880] > > > TX power offset: none > > > Frequency offset: none > > > Supports VAPs: yes PHY name: phy0 > > > > > > Status of radio Node B > > > wlan0-mesh ESSID: "speedtest" > > > Access Point: 04:F0:21:3E:DA:A2 > > > Mode: Master Channel: 48 (5.240 GHz) > > > Tx-Power: 8 dBm Link Quality: 65/70 > > > Signal: -45 dBm Noise: -105 dBm > > > Bit Rate: 6.0 MBit/s > > > Encryption: none > > > Type: nl80211 HW Mode(s): 802.11bgnac > > > Hardware: 168C:003C : [Qualcomm Atheros QCA9880] > > > TX power offset: none > > > Frequency offset: none > > > Supports VAPs: yes PHY name: phy0 > > > > > > > > > Ping between radios > > > > > > root@FlatsEast:~# ping 10.69.0.1 > > > PING 10.69.0.1 (10.69.0.1) 56(84) bytes of data. > > > 64 bytes from 10.69.0.1:
Re: [OpenWrt-Devel] trying to acheive close to 1 Gbps with 802.11ac 3x3 mimo ath10k Compex wle900vx
It looks like Ben was right Further investigation is showing that my Compex WPJ563 (Qualcomm Atheros QCA9563 74Kc MIPS processor, Dragonfly Series 64KB I-Cache and 32 KB D-Cache, targeted to operate at 775MHz 128MB System Memory, 16MB NOR Flash) is the bottleneck. I've tested on the 1Gbps Ethernet link between a laptop with a 1Gbps port and the Compex 1Gbps port and the best throughput I got was 98 Mbps. This was with 3 TCP streams on iperf. I'm still finding it hard to believe that I can't get more speed out of this compex board running openwrt ... top is not showing the CPU go over 30% with iperf when the test is running Thanks for the other advice - I've built a test rig with antennas oriented like a propeller at 120 degrees (I'll share a pic) ... once I sort out the underlying problem or upgrade my router board - I'll test this mimo rig. David On Fri, Oct 5, 2018 at 6:02 PM Ben Greear wrote: > > If you are running iperf on a weak-ish AP CPU, then that will be a bottleneck. > Run iperf on external system through the AP instead. For TCP, use multiple > streams, > and first test with Ethernet mode to make sure iperf and/or CPU is not the > bottleneck > before you move to testing the WiFi interfaces. > > We can get around 900Mbps UDP download throughput on a 'perfect' setup with > 3x3 9880 WLE900VX > systems and the ath10k-ct firmware. > > Forcing MCS rate is not a good idea...tune antennas and/or RF environment so > that you do not have > to force MCS to get highest encoding rate. > > Thanks, > Ben > > On 10/05/2018 04:17 AM, David Johnson wrote: > > I'm trying to see what max speed can be achieved > > > > My setup is as follow: > > Board: Compex WPJ 563 > > Card Compex wle900vx > > OS: Libremesh running openwrt 18.06 > > Driver: ath10k_pci :00:00.0: firmware ver 10.2.4-1.0-00033 api 5 > > features no-p2p,raw-mode,mfp,allows-mesh-bcast crc32 c41417d0 > > Firmware: firmware-5.bin_10.2.4-1.0-00041 > > > > Physical setup > > Connected 3 6dBi antennas on each side to CH0, CH1, CH2 port to > > achieve 3x3 mimo. Distance between 2 devices is 2m on a bench. Set > > power level on each end to 8 dBm to ensure I don't saturate radios > > > > Status of radio Node A > > iwinfo wlan0-mesh info > > wlan0-mesh ESSID: "speedtest" > > Access Point: 04:F0:21:3E:DA:A2 > > Mode: Client Channel: 48 (5.240 GHz) > > Tx-Power: 8 dBm Link Quality: 70/70 > > Signal: -33 dBm Noise: -102 dBm > > Bit Rate: 6.0 MBit/s > > Encryption: none > > Type: nl80211 HW Mode(s): 802.11bgnac > > Hardware: 168C:003C : [Qualcomm Atheros QCA9880] > > TX power offset: none > > Frequency offset: none > > Supports VAPs: yes PHY name: phy0 > > > > Status of radio Node B > > wlan0-mesh ESSID: "speedtest" > > Access Point: 04:F0:21:3E:DA:A2 > > Mode: Master Channel: 48 (5.240 GHz) > > Tx-Power: 8 dBm Link Quality: 65/70 > > Signal: -45 dBm Noise: -105 dBm > > Bit Rate: 6.0 MBit/s > > Encryption: none > > Type: nl80211 HW Mode(s): 802.11bgnac > > Hardware: 168C:003C : [Qualcomm Atheros QCA9880] > > TX power offset: none > > Frequency offset: none > > Supports VAPs: yes PHY name: phy0 > > > > > > Ping between radios > > > > root@FlatsEast:~# ping 10.69.0.1 > > PING 10.69.0.1 (10.69.0.1) 56(84) bytes of data. > > 64 bytes from 10.69.0.1: icmp_req=1 ttl=64 time=1.29 ms > > 64 bytes from 10.69.0.1: icmp_req=2 ttl=64 time=1.22 ms > > 64 bytes from 10.69.0.1: icmp_req=3 ttl=64 time=1.22 ms > > > > Used iperf to set throughput > > Client side > > iperf -c 10.69.0.1 -P 10 -i 5 -t 120 > > > > Server side > > iperf -s > > > > Best performance achieved after waiting for TCP to stabalize: 139 Mbps > > [ 12] 45.0-50.0 sec 11.8 MBytes 19.7 Mbits/sec > > [ 11] 45.0-50.0 sec 7.12 MBytes 12.0 Mbits/sec > > [ 9] 45.0-50.0 sec 8.12 MBytes 13.6 Mbits/sec > > [ 6] 45.0-50.0 sec 7.75 MBytes 13.0 Mbits/sec > > [ 8] 45.0-50.0 sec 8.00 MBytes 13.4 Mbits/sec > > [ 4] 45.0-50.0 sec 7.75 MBytes 13.0 Mbits/sec > > [ 3] 45.0-50.0 sec 7.75 MBytes 13.0 Mbits/sec > > [ 10] 45.0-50.0 sec 6.62 MBytes 11.1 Mbits/sec > > [ 7] 45.0-50.0 sec 7.88 MBytes 13.2 Mbits/sec > > [ 5] 45.0-50.0 sec 9.88 MBytes 16.6 Mbits/sec > > [SUM] 45.0-50.0 sec 82.6 MBytes 139 Mbits/sec > > > > top: > > > > Mem: 47064K used, 78232K free, 188K shrd, 3936K buff, 10564K cached > > CPU: 19% usr 36% sys 0% nic 0% idle 0% io 0% irq 43% sirq > > Load average: 4.59 3.06 1.90 3/74 25823 > > PID PPID USER STAT VSZ %VSZ %CPU COMMAND > > 3 2 root RW 0 0% 32% [ksoftirqd/0] > > 24511 4419 root S33320 27% 22% iperf -c 10.69.0.1 -P 10 -i 5 -t > > 120 > > > > Any suggestions > > > > Regards > > David > > > > -- > Ben Greear > Candela Technologies Inc http://www.candelatech.com
Re: [OpenWrt-Devel] trying to acheive close to 1 Gbps with 802.11ac 3x3 mimo ath10k Compex wle900vx
If you are running iperf on a weak-ish AP CPU, then that will be a bottleneck. Run iperf on external system through the AP instead. For TCP, use multiple streams, and first test with Ethernet mode to make sure iperf and/or CPU is not the bottleneck before you move to testing the WiFi interfaces. We can get around 900Mbps UDP download throughput on a 'perfect' setup with 3x3 9880 WLE900VX systems and the ath10k-ct firmware. Forcing MCS rate is not a good idea...tune antennas and/or RF environment so that you do not have to force MCS to get highest encoding rate. Thanks, Ben On 10/05/2018 04:17 AM, David Johnson wrote: I'm trying to see what max speed can be achieved My setup is as follow: Board: Compex WPJ 563 Card Compex wle900vx OS: Libremesh running openwrt 18.06 Driver: ath10k_pci :00:00.0: firmware ver 10.2.4-1.0-00033 api 5 features no-p2p,raw-mode,mfp,allows-mesh-bcast crc32 c41417d0 Firmware: firmware-5.bin_10.2.4-1.0-00041 Physical setup Connected 3 6dBi antennas on each side to CH0, CH1, CH2 port to achieve 3x3 mimo. Distance between 2 devices is 2m on a bench. Set power level on each end to 8 dBm to ensure I don't saturate radios Status of radio Node A iwinfo wlan0-mesh info wlan0-mesh ESSID: "speedtest" Access Point: 04:F0:21:3E:DA:A2 Mode: Client Channel: 48 (5.240 GHz) Tx-Power: 8 dBm Link Quality: 70/70 Signal: -33 dBm Noise: -102 dBm Bit Rate: 6.0 MBit/s Encryption: none Type: nl80211 HW Mode(s): 802.11bgnac Hardware: 168C:003C : [Qualcomm Atheros QCA9880] TX power offset: none Frequency offset: none Supports VAPs: yes PHY name: phy0 Status of radio Node B wlan0-mesh ESSID: "speedtest" Access Point: 04:F0:21:3E:DA:A2 Mode: Master Channel: 48 (5.240 GHz) Tx-Power: 8 dBm Link Quality: 65/70 Signal: -45 dBm Noise: -105 dBm Bit Rate: 6.0 MBit/s Encryption: none Type: nl80211 HW Mode(s): 802.11bgnac Hardware: 168C:003C : [Qualcomm Atheros QCA9880] TX power offset: none Frequency offset: none Supports VAPs: yes PHY name: phy0 Ping between radios root@FlatsEast:~# ping 10.69.0.1 PING 10.69.0.1 (10.69.0.1) 56(84) bytes of data. 64 bytes from 10.69.0.1: icmp_req=1 ttl=64 time=1.29 ms 64 bytes from 10.69.0.1: icmp_req=2 ttl=64 time=1.22 ms 64 bytes from 10.69.0.1: icmp_req=3 ttl=64 time=1.22 ms Used iperf to set throughput Client side iperf -c 10.69.0.1 -P 10 -i 5 -t 120 Server side iperf -s Best performance achieved after waiting for TCP to stabalize: 139 Mbps [ 12] 45.0-50.0 sec 11.8 MBytes 19.7 Mbits/sec [ 11] 45.0-50.0 sec 7.12 MBytes 12.0 Mbits/sec [ 9] 45.0-50.0 sec 8.12 MBytes 13.6 Mbits/sec [ 6] 45.0-50.0 sec 7.75 MBytes 13.0 Mbits/sec [ 8] 45.0-50.0 sec 8.00 MBytes 13.4 Mbits/sec [ 4] 45.0-50.0 sec 7.75 MBytes 13.0 Mbits/sec [ 3] 45.0-50.0 sec 7.75 MBytes 13.0 Mbits/sec [ 10] 45.0-50.0 sec 6.62 MBytes 11.1 Mbits/sec [ 7] 45.0-50.0 sec 7.88 MBytes 13.2 Mbits/sec [ 5] 45.0-50.0 sec 9.88 MBytes 16.6 Mbits/sec [SUM] 45.0-50.0 sec 82.6 MBytes 139 Mbits/sec top: Mem: 47064K used, 78232K free, 188K shrd, 3936K buff, 10564K cached CPU: 19% usr 36% sys 0% nic 0% idle 0% io 0% irq 43% sirq Load average: 4.59 3.06 1.90 3/74 25823 PID PPID USER STAT VSZ %VSZ %CPU COMMAND 3 2 root RW 0 0% 32% [ksoftirqd/0] 24511 4419 root S33320 27% 22% iperf -c 10.69.0.1 -P 10 -i 5 -t 120 Any suggestions Regards David -- Ben Greear Candela Technologies Inc http://www.candelatech.com ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [OpenWrt-Devel] trying to acheive close to 1 Gbps with 802.11ac 3x3 mimo ath10k Compex wle900vx
I better clarify. the polarization planes would be 60deg apart. The vertical antennas would be physically mounted 120deg apart like a 3 blade airplane propeller. ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [OpenWrt-Devel] trying to acheive close to 1 Gbps with 802.11ac 3x3 mimo ath10k Compex wle900vx
Ditto on channel width, max rate is cut in half with cutting channel width in half. How do you have the antennas oriented? If all are in the same plane or polarization, then attempting to transmit different data on the antennas will directly compete and interfere with all the signals on the same signal polarization. The radio would likely end up sending the same data out all the antennas using lower MCS rates. Orient the antennas, probably need 60deg angles for 3 antennas to minimize the interference between polarization. Align the antennas the same orientation on both ends. Joe AE6XE ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [OpenWrt-Devel] trying to acheive close to 1 Gbps with 802.11ac 3x3 mimo ath10k Compex wle900vx
A few other parameters I left out Channel:48 Bandwidth: 80 MHz (max allowed for this compex card using Qualcomm-Atheros QCA9880) As this is an indoor lab bench test over 2 metres - I don't think the channel interference should matter much (reported noise is -102 dBm) I tried: iw wlan0-mesh set bitrates vht-mcs-5 3:0-9 (i.e. force to 11ac mode with 3 streams) which sometimes caused the kernel to crash ... logs showed: ath10k_pci :00:00.0: firmware crashed! The ath10k-ct driver didn't crash with this command but it actually made the speed worse On Fri, Oct 5, 2018 at 1:17 PM David Johnson wrote: > > I'm trying to see what max speed can be achieved > > My setup is as follow: > Board: Compex WPJ 563 > Card Compex wle900vx > OS: Libremesh running openwrt 18.06 > Driver: ath10k_pci :00:00.0: firmware ver 10.2.4-1.0-00033 api 5 > features no-p2p,raw-mode,mfp,allows-mesh-bcast crc32 c41417d0 > Firmware: firmware-5.bin_10.2.4-1.0-00041 > > Physical setup > Connected 3 6dBi antennas on each side to CH0, CH1, CH2 port to > achieve 3x3 mimo. Distance between 2 devices is 2m on a bench. Set > power level on each end to 8 dBm to ensure I don't saturate radios > > Status of radio Node A > iwinfo wlan0-mesh info > wlan0-mesh ESSID: "speedtest" > Access Point: 04:F0:21:3E:DA:A2 > Mode: Client Channel: 48 (5.240 GHz) > Tx-Power: 8 dBm Link Quality: 70/70 > Signal: -33 dBm Noise: -102 dBm > Bit Rate: 6.0 MBit/s > Encryption: none > Type: nl80211 HW Mode(s): 802.11bgnac > Hardware: 168C:003C : [Qualcomm Atheros QCA9880] > TX power offset: none > Frequency offset: none > Supports VAPs: yes PHY name: phy0 > > Status of radio Node B > wlan0-mesh ESSID: "speedtest" > Access Point: 04:F0:21:3E:DA:A2 > Mode: Master Channel: 48 (5.240 GHz) > Tx-Power: 8 dBm Link Quality: 65/70 > Signal: -45 dBm Noise: -105 dBm > Bit Rate: 6.0 MBit/s > Encryption: none > Type: nl80211 HW Mode(s): 802.11bgnac > Hardware: 168C:003C : [Qualcomm Atheros QCA9880] > TX power offset: none > Frequency offset: none > Supports VAPs: yes PHY name: phy0 > > > Ping between radios > > root@FlatsEast:~# ping 10.69.0.1 > PING 10.69.0.1 (10.69.0.1) 56(84) bytes of data. > 64 bytes from 10.69.0.1: icmp_req=1 ttl=64 time=1.29 ms > 64 bytes from 10.69.0.1: icmp_req=2 ttl=64 time=1.22 ms > 64 bytes from 10.69.0.1: icmp_req=3 ttl=64 time=1.22 ms > > Used iperf to set throughput > Client side > iperf -c 10.69.0.1 -P 10 -i 5 -t 120 > > Server side > iperf -s > > Best performance achieved after waiting for TCP to stabalize: 139 Mbps > [ 12] 45.0-50.0 sec 11.8 MBytes 19.7 Mbits/sec > [ 11] 45.0-50.0 sec 7.12 MBytes 12.0 Mbits/sec > [ 9] 45.0-50.0 sec 8.12 MBytes 13.6 Mbits/sec > [ 6] 45.0-50.0 sec 7.75 MBytes 13.0 Mbits/sec > [ 8] 45.0-50.0 sec 8.00 MBytes 13.4 Mbits/sec > [ 4] 45.0-50.0 sec 7.75 MBytes 13.0 Mbits/sec > [ 3] 45.0-50.0 sec 7.75 MBytes 13.0 Mbits/sec > [ 10] 45.0-50.0 sec 6.62 MBytes 11.1 Mbits/sec > [ 7] 45.0-50.0 sec 7.88 MBytes 13.2 Mbits/sec > [ 5] 45.0-50.0 sec 9.88 MBytes 16.6 Mbits/sec > [SUM] 45.0-50.0 sec 82.6 MBytes 139 Mbits/sec > > top: > > Mem: 47064K used, 78232K free, 188K shrd, 3936K buff, 10564K cached > CPU: 19% usr 36% sys 0% nic 0% idle 0% io 0% irq 43% sirq > Load average: 4.59 3.06 1.90 3/74 25823 > PID PPID USER STAT VSZ %VSZ %CPU COMMAND > 3 2 root RW 0 0% 32% [ksoftirqd/0] > 24511 4419 root S33320 27% 22% iperf -c 10.69.0.1 -P 10 -i 5 -t 120 > > Any suggestions > > Regards > David > > -- > Dr David Johnson > Director, Ammbr Research Labs South Africa > Part of the AmmbrTech Group > http://www.ammbrtech.com > > Adjunct Senior Lecturer > ICT4D, Computer Science Department > University of Cape Town > https://people.cs.uct.ac.za/~djohnson/ ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel
Re: [OpenWrt-Devel] trying to acheive close to 1 Gbps with 802.11ac 3x3 mimo ath10k Compex wle900vx
> Any suggestions What have you tried in terms of channel bandwidth? Is this with 20MHz? 802.11ac goes up to 160 MHz, maybe you should try that. Chan 48 is also pretty low in the spectrum, maybe try some of the higher DFS channels, at least in dense city environments these tend to me much less used by other wifis. - Arne ___ openwrt-devel mailing list openwrt-devel@lists.openwrt.org https://lists.openwrt.org/mailman/listinfo/openwrt-devel